fix: Bedrock: Mantle models template api on ${AWS_REGION}, but aren't served in every region

Co-authored-by: opencode-agent[bot] <opencode-agent[bot]@users.noreply.github.com>

.github/workflows/issue-fixer.yml

@@ -18,9 +18,7 @@ jobs:
     if: >-
       github.repository == 'anomalyco/models.dev'
       && !contains(github.event.issue.labels.*.name, 'provider:openai')
-      && !contains(github.event.issue.labels.*.name, 'provider:pioneer')
       && github.event.client_payload.provider != 'openai'
-      && github.event.client_payload.provider != 'pioneer'
     runs-on: ubuntu-latest
     env:
       GH_TOKEN: ${{ github.token }}

.github/workflows/pr-reviewer.yml

@@ -7,6 +7,7 @@ on:
 
 permissions:
   contents: read
+  issues: write
   pull-requests: write
 
 concurrency:
@@ -22,6 +23,19 @@ jobs:
     runs-on: ubuntu-latest
 
     steps:
+      - name: Clear ready label
+        env:
+          GH_TOKEN: ${{ github.token }}
+          PR_NUMBER: ${{ github.event.pull_request.number }}
+          READY_LABEL: "reviewer: ready"
+        run: |
+          set -euo pipefail
+          gh label create "$READY_LABEL" --repo "$GITHUB_REPOSITORY" --color "0E8A16" --description "Automated review found no actionable items" --force
+          labels="$(gh pr view "$PR_NUMBER" --repo "$GITHUB_REPOSITORY" --json labels --jq '.labels[].name')"
+          if grep -Fxq "$READY_LABEL" <<< "$labels"; then
+            gh pr edit "$PR_NUMBER" --repo "$GITHUB_REPOSITORY" --remove-label "$READY_LABEL"
+          fi
+
       - name: Checkout trusted base revision
         uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5
         with:
@@ -53,15 +67,19 @@ jobs:
       - name: Run pull request reviewer
         env:
           OPENCODE_API_KEY: ${{ secrets.OPENCODE_API_KEY }}
-          OPENCODE_PERMISSION: '{"*":"deny","read":"allow","glob":"allow","grep":"allow","external_directory":"deny"}'
+          OPENCODE_PERMISSION: '{"*":"deny","read":"allow","glob":"allow","grep":"allow","mark-pr-ready":"allow","external_directory":"deny"}'
         run: |
           set -euo pipefail
           EVENTS_FILE="$RUNNER_TEMP/pr-reviewer-events.jsonl"
           RESPONSE_FILE="$RUNNER_TEMP/pr-reviewer-response.md"
+          PR_REVIEW_READY_FILE="$RUNNER_TEMP/pr-reviewer-ready"
           echo "RESPONSE_FILE=$RESPONSE_FILE" >> "$GITHUB_ENV"
+          echo "PR_REVIEW_READY_FILE=$PR_REVIEW_READY_FILE" >> "$GITHUB_ENV"
+          export PR_REVIEW_READY_FILE
+          rm -f "$PR_REVIEW_READY_FILE"
 
           opencode run --agent pr-reviewer -m opencode/grok-4.5 --format json <<'EOF' | tee "$EVENTS_FILE"
-          Review this pull request using the trusted reviewer instructions. Start with `.pr-review/pull-request.json`, `.pr-review/diff.patch`, `AGENTS.md`, and the contributing guidance in `README.md`. Read `sync.md`, the reasoning-options audit guide, schema code, and nearby base-revision files when relevant to the changed files. Use only the read, glob, and grep tools. Return only the final review comment in the agent's required output format. Never include progress narration or passed-check summaries.
+          Review this pull request using the trusted reviewer instructions. Start with `.pr-review/pull-request.json`, `.pr-review/diff.patch`, `AGENTS.md`, and the contributing guidance in `README.md`. Read `sync.md`, the reasoning-options audit guide, schema code, and nearby base-revision files when relevant to the changed files. Use only the read, glob, grep, and mark-pr-ready tools. Return only the final review comment in the agent's required output format. Never include progress narration or passed-check summaries.
           EOF
 
           if ! jq -ers 'map(select(.type == "text") | .part.text) | last | select(length > 0)' "$EVENTS_FILE" > "$RESPONSE_FILE"; then
@@ -73,4 +91,9 @@ jobs:
         env:
           GH_TOKEN: ${{ github.token }}
           PR_NUMBER: ${{ github.event.pull_request.number }}
-        run: gh pr comment "$PR_NUMBER" --repo "$GITHUB_REPOSITORY" --body-file "$RESPONSE_FILE"
+          READY_LABEL: "reviewer: ready"
+        run: |
+          gh pr comment "$PR_NUMBER" --repo "$GITHUB_REPOSITORY" --body-file "$RESPONSE_FILE"
+          if [[ -f "$PR_REVIEW_READY_FILE" ]]; then
+            gh pr edit "$PR_NUMBER" --repo "$GITHUB_REPOSITORY" --add-label "$READY_LABEL"
+          fi

.opencode/agent/pr-reviewer.md

@@ -12,6 +12,7 @@ permission:
     "*.env.*": deny
   glob: allow
   grep: allow
+  mark-pr-ready: allow
   external_directory: deny
 ---
 
@@ -65,6 +66,8 @@ Focus only on actionable problems introduced by the pull request:
 
 Do not report style preferences, speculative concerns, pre-existing problems, or bare schema errors that validation will identify without useful explanation. Do not invent requirements from neighboring files when provider behavior is intentionally different. Do not claim to have run commands, opened links, or performed validation. Do not edit files or attempt to post comments yourself.
 
+Use `mark-pr-ready` only after completing the review and determining there are no action items. Never use it when returning one or more action items.
+
 Every finding must be an action item: the author must need to change something, verify a specific fact, or provide missing evidence. Do not list checks that passed or general observations. If you find action items, list them in severity order and return exactly this structure:
 
 ```markdown
@@ -74,6 +77,6 @@ Every finding must be an action item: the author must need to change something,
 
 Use `violation` only when the change demonstrably breaks a repository requirement or expected behavior. Use `possible mistake` when the diff provides concrete contradictory or suspicious evidence but external facts must be verified. Use `critical`, `high`, `medium`, or `low` for severity. Reference a changed line whenever possible and keep each action item concise.
 
-If there are no action items, respond with exactly the following text and nothing else. Do not explain what you checked or why it passed:
+If there are no action items, call `mark-pr-ready`, then respond with exactly the following text and nothing else. Do not explain what you checked or why it passed:
 
 `No actionable findings.`

.opencode/opencode.jsonc

@@ -0,0 +1,6 @@
+{
+  "$schema": "https://opencode.ai/config.json",
+  "permission": {
+    "mark-pr-ready": "deny"
+  }
+}

.opencode/tool/mark-pr-ready.ts

@@ -0,0 +1,16 @@
+import { writeFile } from "node:fs/promises"
+import { tool } from "@opencode-ai/plugin"
+
+export default tool({
+  description: "Mark the current pull request as ready after completing a review with no actionable findings.",
+  args: {},
+  async execute(_args, context) {
+    if (context.agent !== "pr-reviewer") throw new Error("This tool is only available to the pr-reviewer agent")
+
+    const readyFile = process.env.PR_REVIEW_READY_FILE
+    if (!readyFile) throw new Error("PR_REVIEW_READY_FILE is not configured")
+
+    await writeFile(readyFile, "")
+    return "Pull request marked ready."
+  },
+})

labs/arcee-ai/lab.toml

@@ -0,0 +1 @@
+description = "Arcee AI develops open-weight language models focused on efficient reasoning, tool use, and deployable intelligence."

labs/arcee-ai/logo.svg

@@ -0,0 +1 @@
+<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24" fill="currentColor" fill-rule="evenodd"><path d="M13.236 2.377 2.751 20.493H0L11.863 0l1.373 2.377zm3.554 6.156-9.606 11.96H4.13L15.511 6.32l1.279 2.212zm6.908 11.96H14.05l8.406-2.151 1.242 2.15zm-3.42-5.922-7.843 5.92H8.482l10.597-7.997 1.2 2.077z"/></svg>

models/aisingapore/gemma-sea-lion-v4-27b-it.toml

@@ -0,0 +1,22 @@
+name = "Gemma-SEA-LION-v4-27B-IT"
+description = "Gemma 3 27B tuned by AI Singapore for Southeast Asian languages and instruction following"
+family = "gemma"
+release_date = "2025-09-23"
+last_updated = "2025-09-23"
+attachment = false
+reasoning = false
+temperature = true
+tool_call = false
+open_weights = true
+
+[limit]
+context = 128_000
+output = 128_000
+
+[modalities]
+input = ["text"]
+output = ["text"]
+
+[[weights]]
+label = "Hugging Face"
+url = "https://huggingface.co/aisingapore/Gemma-SEA-LION-v4-27B-IT"

models/alibaba/qwen2.5-coder-32b-instruct.toml

@@ -0,0 +1,22 @@
+name = "Qwen2.5-Coder-32B-Instruct"
+description = "Open coding-focused Qwen model for code generation, repair, and repository reasoning"
+family = "qwen"
+release_date = "2024-11-12"
+last_updated = "2024-11-12"
+attachment = false
+reasoning = false
+temperature = true
+tool_call = true
+open_weights = true
+
+[limit]
+context = 131_072
+output = 8_192
+
+[modalities]
+input = ["text"]
+output = ["text"]
+
+[[weights]]
+label = "Hugging Face"
+url = "https://huggingface.co/Qwen/Qwen2.5-Coder-32B-Instruct"

models/alibaba/qwen3-30b-a3b.toml

@@ -0,0 +1,22 @@
+name = "Qwen3 30B A3B"
+description = "Sparse MoE Qwen model with 3B active parameters for efficient chat and reasoning"
+family = "qwen"
+release_date = "2025-04-28"
+last_updated = "2025-04-28"
+attachment = false
+reasoning = true
+temperature = true
+tool_call = true
+open_weights = true
+
+[limit]
+context = 131_072
+output = 16_384
+
+[modalities]
+input = ["text"]
+output = ["text"]
+
+[[weights]]
+label = "Hugging Face"
+url = "https://huggingface.co/Qwen/Qwen3-30B-A3B"

models/alibaba/qwen3-coder-next.toml

@@ -0,0 +1,27 @@
+# https://qwen.ai/blog?id=qwen3-coder-next
+# https://huggingface.co/Qwen/Qwen3-Coder-Next
+# https://www.qwencloud.com/models/qwen3-coder-next
+name = "Qwen3 Coder Next"
+description = "Open-weight Qwen coding model for agents, repository edits, and multi-turn tool use"
+family = "qwen"
+release_date = "2026-02-03"
+last_updated = "2026-02-03"
+attachment = false
+reasoning = false
+temperature = true
+tool_call = true
+structured_output = true
+knowledge = "2025-09"
+open_weights = true
+
+[limit]
+context = 262_144
+output = 65_536
+
+[modalities]
+input = ["text"]
+output = ["text"]
+
+[[weights]]
+label = "Hugging Face"
+url = "https://huggingface.co/Qwen/Qwen3-Coder-Next"

models/alibaba/qwen3.5-flash.toml

@@ -0,0 +1,22 @@
+# https://help.aliyun.com/en/model-studio/qwen3-5-flash
+# https://www.alibabacloud.com/help/en/model-studio/deep-thinking
+name = "Qwen3.5 Flash"
+description = "Qwen vision-language model for visual reasoning, documents, and agent tasks"
+family = "qwen"
+release_date = "2026-02-23"
+last_updated = "2026-02-23"
+attachment = true
+reasoning = true
+temperature = true
+tool_call = true
+structured_output = true
+open_weights = false
+
+
+[limit]
+context = 1_000_000
+output = 65_536
+
+[modalities]
+input = ["text", "image", "video"]
+output = ["text"]

models/alibaba/qwen3.8-27b.toml

@@ -0,0 +1,36 @@
+# Sources (accessed 2026-08-15):
+# https://huggingface.co/Qwen/Qwen3.8-27B
+# https://huggingface.co/api/models/Qwen/Qwen3.8-27B
+# https://qwen.ai/blog?id=qwen3.8
+# Hub lastModified 2026-08-14T15:00:01Z is the open-weight drop.
+# Do not use Hub createdAt 2026-08-05 (staged countdown page).
+
+name = "Qwen3.8 27B"
+description = "Dense 27B vision-language model for coding, agent tasks, and image and video understanding"
+family = "qwen"
+release_date = "2026-08-14"
+last_updated = "2026-08-14"
+attachment = true
+reasoning = true
+temperature = true
+tool_call = true
+structured_output = true
+open_weights = true
+
+[limit]
+context = 262_144
+output = 32_768
+
+[modalities]
+input = ["text", "image", "video"]
+output = ["text"]
+
+[[weights]]
+label = "Hugging Face"
+url = "https://huggingface.co/Qwen/Qwen3.8-27B"
+
+[[benchmarks]]
+name = "SWE-bench Pro"
+score = 61.7
+metric = "resolved"
+source = "https://huggingface.co/Qwen/Qwen3.8-27B"

models/alibaba/qwen3.8-max.toml

@@ -1,6 +1,10 @@
-# Sources (accessed 2026-08-03):
+# Sources (accessed 2026-08-06):
 # https://www.qwencloud.com/models/qwen3.8-max
 # https://www.qianwenai.com/models/qwen3.8-max
+# https://help.aliyun.com/zh/model-studio/qwen3-8-max
+# https://www.alibabacloud.com/help/en/model-studio/qwen3-8-max
+# https://help.aliyun.com/zh/model-studio/pdf-understanding
+# https://platform.qianwenai.com/docs/developer-guides/tool-calling/pdf-understanding
 # https://docs.qwencloud.com/token-plan/personal/token-plan-personal-overview
 # https://help.aliyun.com/zh/model-studio/token-plan-personal-overview
 # https://help.aliyun.com/en/model-studio/token-plan-personal-overview
@@ -9,6 +13,10 @@
 # https://docs.qwencloud.com/developer-guides/clients-and-developer-tools/opencode
 # https://platform.qianwenai.com/docs/developer-guides/clients-and-developer-tools/opencode
 # https://qwen.ai/blog?id=qwen3.8
+# PDF input: Model Studio / 千问AI docs list only qwen3.8-max under PDF理解
+# (type:file / file_url|file_data). Model pages list Image/Text/Video badges
+# and separately list PDF理解 as a Completions built-in tool. Beijing-region
+# availability note on help.aliyun.com; lab capability still includes pdf.
 
 name = "Qwen3.8 Max"
 description = "2.4-trillion-parameter MoE flagship for coding, professional work, multimodal understanding, and long-horizon agentic workflows"
@@ -26,5 +34,5 @@ context = 1_000_000
 output = 131_072
 
 [modalities]
-input = ["text", "image", "video"]
+input = ["text", "image", "video", "pdf"]
 output = ["text"]

models/alibaba/qwq-32b.toml

@@ -0,0 +1,23 @@
+name = "QwQ 32B"
+description = "Open reasoning model from the Qwen team for math, coding, and step-by-step problem solving"
+family = "qwen"
+release_date = "2025-03-05"
+last_updated = "2025-03-05"
+attachment = false
+reasoning = true
+temperature = true
+tool_call = true
+knowledge = "2024-04"
+open_weights = true
+
+[limit]
+context = 131_072
+output = 8_192
+
+[modalities]
+input = ["text"]
+output = ["text"]
+
+[[weights]]
+label = "Hugging Face"
+url = "https://huggingface.co/Qwen/QwQ-32B"

models/anthropic/claude-mythos-5.toml

@@ -0,0 +1,23 @@
+# Sources:
+# https://platform.claude.com/docs/en/about-claude/models/introducing-claude-fable-5-and-claude-mythos-5
+# https://www.anthropic.com/claude/mythos
+name = "Claude Mythos 5"
+description = "Restricted Claude model for advanced cybersecurity and biology research workflows"
+family = "claude-mythos"
+release_date = "2026-06-09"
+last_updated = "2026-06-09"
+attachment = true
+reasoning = true
+temperature = false
+tool_call = true
+structured_output = true
+knowledge = "2026-01-31"
+open_weights = false
+
+[limit]
+context = 1_000_000
+output = 128_000
+
+[modalities]
+input = ["text", "image", "pdf"]
+output = ["text"]

models/arcee-ai/trinity-large-preview.toml

@@ -0,0 +1,40 @@
+# Source: https://huggingface.co/arcee-ai/Trinity-Large-Preview
+name = "Trinity Large Preview"
+description = "Lightly post-trained 398B MoE chat model for creative work, long-context prompts, and tool-using agents"
+family = "trinity"
+release_date = "2026-01-27"
+last_updated = "2026-05-28"
+attachment = false
+reasoning = false
+temperature = true
+tool_call = true
+open_weights = true
+license = "OpenMDW-1.1"
+
+[limit]
+context = 524_288
+output = 262_144
+
+[modalities]
+input = ["text"]
+output = ["text"]
+
+[[weights]]
+label = "Hugging Face"
+url = "https://huggingface.co/arcee-ai/Trinity-Large-Preview"
+format = "safetensors"
+
+[[links]]
+label = "Model card"
+url = "https://huggingface.co/arcee-ai/Trinity-Large-Preview"
+type = "model_card"
+
+[[links]]
+label = "Announcement"
+url = "https://www.arcee.ai/blog/trinity-large"
+type = "announcement"
+
+[[links]]
+label = "License"
+url = "https://huggingface.co/arcee-ai/Trinity-Large-Preview/blob/main/LICENSE"
+type = "license"

models/arcee-ai/trinity-large-thinking.toml

@@ -0,0 +1,40 @@
+# Source: https://huggingface.co/arcee-ai/Trinity-Large-Thinking
+name = "Trinity Large Thinking"
+description = "Reasoning-optimized 398B MoE agent model with extended thinking for long-horizon and multi-turn tool use"
+family = "trinity"
+release_date = "2026-04-01"
+last_updated = "2026-05-28"
+attachment = false
+reasoning = true
+temperature = true
+tool_call = true
+open_weights = true
+license = "OpenMDW-1.1"
+
+[limit]
+context = 524_288
+output = 262_144
+
+[modalities]
+input = ["text"]
+output = ["text"]
+
+[[weights]]
+label = "Hugging Face"
+url = "https://huggingface.co/arcee-ai/Trinity-Large-Thinking"
+format = "safetensors"
+
+[[links]]
+label = "Model card"
+url = "https://huggingface.co/arcee-ai/Trinity-Large-Thinking"
+type = "model_card"
+
+[[links]]
+label = "Announcement"
+url = "https://www.arcee.ai/blog/trinity-large-thinking"
+type = "announcement"
+
+[[links]]
+label = "License"
+url = "https://huggingface.co/arcee-ai/Trinity-Large-Thinking/blob/main/LICENSE"
+type = "license"

models/arcee-ai/trinity-mini.toml

@@ -0,0 +1,40 @@
+# Source: https://huggingface.co/arcee-ai/Trinity-Mini
+name = "Trinity Mini"
+description = "Reasoning-tuned 26B MoE model with 3B active parameters for agents, tools, and multi-step workloads"
+family = "trinity"
+release_date = "2025-12-01"
+last_updated = "2026-05-28"
+attachment = false
+reasoning = true
+temperature = true
+tool_call = true
+open_weights = true
+license = "OpenMDW-1.1"
+
+[limit]
+context = 131_072
+output = 131_072
+
+[modalities]
+input = ["text"]
+output = ["text"]
+
+[[weights]]
+label = "Hugging Face"
+url = "https://huggingface.co/arcee-ai/Trinity-Mini"
+format = "safetensors"
+
+[[links]]
+label = "Model card"
+url = "https://huggingface.co/arcee-ai/Trinity-Mini"
+type = "model_card"
+
+[[links]]
+label = "Announcement"
+url = "https://www.arcee.ai/blog/the-trinity-manifesto"
+type = "announcement"
+
+[[links]]
+label = "License"
+url = "https://huggingface.co/arcee-ai/Trinity-Mini/blob/main/LICENSE"
+type = "license"

models/arcee-ai/trinity-nano-preview.toml

@@ -0,0 +1,40 @@
+# Source: https://huggingface.co/arcee-ai/Trinity-Nano-Preview
+name = "Trinity Nano Preview"
+description = "Experimental chat-tuned 6B MoE model with 1B active parameters for low-resource chat and instruction following"
+family = "trinity"
+release_date = "2025-12-01"
+last_updated = "2026-05-28"
+attachment = false
+reasoning = false
+temperature = true
+tool_call = true
+open_weights = true
+license = "OpenMDW-1.1"
+
+[limit]
+context = 131_072
+output = 131_072
+
+[modalities]
+input = ["text"]
+output = ["text"]
+
+[[weights]]
+label = "Hugging Face"
+url = "https://huggingface.co/arcee-ai/Trinity-Nano-Preview"
+format = "safetensors"
+
+[[links]]
+label = "Model card"
+url = "https://huggingface.co/arcee-ai/Trinity-Nano-Preview"
+type = "model_card"
+
+[[links]]
+label = "Announcement"
+url = "https://www.arcee.ai/blog/the-trinity-manifesto"
+type = "announcement"
+
+[[links]]
+label = "License"
+url = "https://huggingface.co/arcee-ai/Trinity-Nano-Preview/blob/main/LICENSE"
+type = "license"

models/bytedance-seed/seed-1-6-flash.toml

@@ -0,0 +1,21 @@
+# Sources (accessed 2026-08-14):
+# - https://seed.bytedance.com/en/seed2
+# - https://www.volcengine.com/docs/82379/1330310
+name = "Seed 1.6 Flash"
+description = "Low-latency ByteDance Seed model for high-throughput chat, extraction, and lightweight tool use"
+family = "seed"
+release_date = "2025-08-28"
+last_updated = "2025-08-28"
+attachment = false
+reasoning = false
+temperature = true
+tool_call = true
+open_weights = false
+
+[limit]
+context = 256_000
+output = 32_000
+
+[modalities]
+input = ["text"]
+output = ["text"]

models/bytedance-seed/seed-1-6-vision.toml

@@ -0,0 +1,21 @@
+# Sources (accessed 2026-08-14):
+# - https://seed.bytedance.com/en/seed2
+# - https://www.volcengine.com/docs/82379/1330310
+name = "Seed 1.6 Vision"
+description = "ByteDance Seed multimodal model for image understanding, visual reasoning, and tool-assisted tasks"
+family = "seed"
+release_date = "2025-08-15"
+last_updated = "2025-08-15"
+attachment = true
+reasoning = false
+temperature = true
+tool_call = true
+open_weights = false
+
+[limit]
+context = 256_000
+output = 32_000
+
+[modalities]
+input = ["text", "image"]
+output = ["text"]

models/bytedance-seed/seed-1-6.toml

@@ -0,0 +1,21 @@
+# Sources (accessed 2026-08-14):
+# - https://seed.bytedance.com/en/seed2
+# - https://www.volcengine.com/docs/82379/1330310
+name = "Seed 1.6"
+description = "ByteDance Seed model for long-context reasoning, instruction following, and tool-assisted tasks"
+family = "seed"
+release_date = "2025-10-15"
+last_updated = "2025-10-15"
+attachment = false
+reasoning = true
+temperature = true
+tool_call = true
+open_weights = false
+
+[limit]
+context = 256_000
+output = 64_000
+
+[modalities]
+input = ["text"]
+output = ["text"]

models/bytedance-seed/seed-1-8.toml

@@ -0,0 +1,21 @@
+# Sources (accessed 2026-08-14):
+# - https://seed.bytedance.com/en/seed2
+# - https://www.volcengine.com/docs/82379/1330310
+name = "Seed 1.8"
+description = "ByteDance Seed model for multimodal reasoning, long-context analysis, and agent workflows"
+family = "seed"
+release_date = "2025-12-28"
+last_updated = "2025-12-28"
+attachment = false
+reasoning = true
+temperature = true
+tool_call = true
+open_weights = false
+
+[limit]
+context = 256_000
+output = 64_000
+
+[modalities]
+input = ["text"]
+output = ["text"]

models/bytedance-seed/seed-2.0-code.toml

@@ -0,0 +1,23 @@
+# Sources (accessed 2026-08-11):
+# - https://seed.bytedance.com/en/blog/seed-2-0-official-launch
+# - https://seed.bytedance.com/en/seed2
+# - https://www.volcengine.com/docs/82379/1330310
+name = "Seed 2.0 Code"
+description = "ByteDance Seed coding model for multimodal software engineering and long-running agents"
+family = "seed"
+release_date = "2026-02-14"
+last_updated = "2026-02-14"
+attachment = true
+reasoning = true
+temperature = true
+tool_call = true
+structured_output = true
+open_weights = false
+
+[limit]
+context = 262_144
+output = 131_072
+
+[modalities]
+input = ["text", "image", "video"]
+output = ["text"]

models/bytedance-seed/seed-2.0-lite.toml

@@ -0,0 +1,21 @@
+# Sources (accessed 2026-08-14):
+# - https://seed.bytedance.com/en/seed2
+# - https://www.volcengine.com/docs/82379/1330310
+name = "Seed 2.0 Lite"
+description = "Cost-efficient ByteDance Seed 2.0 model for production chat, analysis, and structured generation"
+family = "seed"
+release_date = "2026-02-14"
+last_updated = "2026-02-14"
+attachment = true
+reasoning = true
+temperature = true
+tool_call = true
+open_weights = false
+
+[limit]
+context = 256_000
+output = 32_000
+
+[modalities]
+input = ["text", "image", "video"]
+output = ["text"]

models/bytedance-seed/seed-2.0-mini.toml

@@ -0,0 +1,21 @@
+# Sources (accessed 2026-08-14):
+# - https://seed.bytedance.com/en/seed2
+# - https://www.volcengine.com/docs/82379/1330310
+name = "Seed 2.0 Mini"
+description = "Lightweight ByteDance Seed 2.0 model for low-latency multimodal reasoning and high-volume tasks"
+family = "seed"
+release_date = "2026-02-14"
+last_updated = "2026-02-14"
+attachment = true
+reasoning = true
+temperature = true
+tool_call = true
+open_weights = false
+
+[limit]
+context = 256_000
+output = 32_000
+
+[modalities]
+input = ["text", "image", "video"]
+output = ["text"]

models/bytedance-seed/seed-2.0-pro.toml

@@ -0,0 +1,21 @@
+# Sources (accessed 2026-08-14):
+# - https://seed.bytedance.com/en/seed2
+# - https://www.volcengine.com/docs/82379/1330310
+name = "Seed 2.0 Pro"
+description = "Flagship ByteDance Seed 2.0 model for complex multimodal reasoning and long-horizon agent workflows"
+family = "seed"
+release_date = "2026-02-14"
+last_updated = "2026-02-14"
+attachment = true
+reasoning = true
+temperature = true
+tool_call = true
+open_weights = false
+
+[limit]
+context = 256_000
+output = 128_000
+
+[modalities]
+input = ["text", "image", "video"]
+output = ["text"]

models/bytedance-seed/seed-2.1-pro.toml

@@ -0,0 +1,21 @@
+# Sources (accessed 2026-08-14):
+# - https://seed.bytedance.com/en/seed2
+# - https://www.volcengine.com/docs/82379/1330310
+name = "Seed 2.1 Pro"
+description = "Flagship ByteDance Seed 2.1 model for complex multimodal reasoning, coding, and agents"
+family = "seed"
+release_date = "2026-06-23"
+last_updated = "2026-06-23"
+attachment = true
+reasoning = true
+temperature = true
+tool_call = true
+open_weights = false
+
+[limit]
+context = 256_000
+output = 256_000
+
+[modalities]
+input = ["text", "image", "video"]
+output = ["text"]

models/bytedance-seed/seed-2.1-turbo.toml

@@ -0,0 +1,21 @@
+# Sources (accessed 2026-08-14):
+# - https://seed.bytedance.com/en/seed2
+# - https://www.volcengine.com/docs/82379/1330310
+name = "Seed 2.1 Turbo"
+description = "Faster ByteDance Seed 2.1 model for multimodal reasoning and latency-sensitive agent workflows"
+family = "seed"
+release_date = "2026-06-23"
+last_updated = "2026-06-23"
+attachment = true
+reasoning = true
+temperature = true
+tool_call = true
+open_weights = false
+
+[limit]
+context = 256_000
+output = 256_000
+
+[modalities]
+input = ["text", "image", "video"]
+output = ["text"]

models/bytedance-seed/seed-character.toml

@@ -0,0 +1,21 @@
+# Sources (accessed 2026-08-14):
+# - https://seed.bytedance.com/en/seed2
+# - https://www.volcengine.com/docs/82379/1330310
+name = "Seed Character"
+description = "ByteDance Seed model optimized for character-driven dialogue and consistent conversational behavior"
+family = "seed"
+release_date = "2026-06-23"
+last_updated = "2026-06-23"
+attachment = true
+reasoning = true
+temperature = true
+tool_call = true
+open_weights = false
+
+[limit]
+context = 256_000
+output = 256_000
+
+[modalities]
+input = ["text", "image", "video"]
+output = ["text"]

models/bytedance-seed/seed-evolving.toml

@@ -0,0 +1,21 @@
+# Sources (accessed 2026-08-14):
+# - https://seed.bytedance.com/en/seed2
+# - https://www.volcengine.com/docs/82379/1330310
+name = "Seed Evolving"
+description = "Rolling ByteDance Seed model for rapidly updated reasoning, coding, and agent capabilities"
+family = "seed"
+release_date = "2026-06-23"
+last_updated = "2026-06-23"
+attachment = true
+reasoning = true
+temperature = true
+tool_call = true
+open_weights = false
+
+[limit]
+context = 256_000
+output = 256_000
+
+[modalities]
+input = ["text", "image", "video"]
+output = ["text"]

models/deepseek/deepseek-ocr-2.toml

@@ -0,0 +1,16 @@
+name = "DeepSeek OCR 2"
+description = "High-accuracy OCR model for extracting text from documents, screenshots, receipts, and natural scenes"
+release_date = "2026-01-27"
+last_updated = "2026-01-27"
+attachment = true
+reasoning = false
+tool_call = false
+open_weights = true
+
+[limit]
+context = 8_192
+output = 8_192
+
+[modalities]
+input = ["text", "image"]
+output = ["text"]

models/deepseek/deepseek-r1-distill-qwen-32b.toml

@@ -0,0 +1,22 @@
+name = "DeepSeek-R1-Distill-Qwen-32B"
+description = "R1 reasoning distilled into Qwen 2.5 32B for efficient open-weight step-by-step problem solving"
+family = "deepseek-thinking"
+release_date = "2025-01-20"
+last_updated = "2025-01-20"
+attachment = false
+reasoning = true
+temperature = true
+tool_call = false
+open_weights = true
+
+[limit]
+context = 131_072
+output = 32_768
+
+[modalities]
+input = ["text"]
+output = ["text"]
+
+[[weights]]
+label = "Hugging Face"
+url = "https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-32B"

models/deepseek/deepseek-v3-0324.toml

@@ -0,0 +1,23 @@
+name = "DeepSeek V3 0324"
+description = "March 2025 checkpoint of DeepSeek-V3 with improved reasoning and coding"
+family = "deepseek"
+release_date = "2025-03-24"
+last_updated = "2025-03-24"
+attachment = false
+reasoning = false
+temperature = true
+tool_call = true
+open_weights = true
+
+[limit]
+context = 163_840
+output = 163_840
+
+[modalities]
+input = ["text"]
+output = ["text"]
+
+[[weights]]
+label = "Model weights"
+url = "https://huggingface.co/deepseek-ai/DeepSeek-V3-0324"
+format = "safetensors"

models/deepseek/deepseek-v3.2.toml

@@ -0,0 +1,27 @@
+# https://api-docs.deepseek.com/news/news251201
+# https://huggingface.co/deepseek-ai/DeepSeek-V3.2
+name = "DeepSeek V3.2"
+description = "Hybrid-reasoning DeepSeek model with thinking and non-thinking modes, sparse attention, and tool-use"
+family = "deepseek"
+release_date = "2025-12-01"
+last_updated = "2025-12-01"
+attachment = false
+reasoning = true
+temperature = true
+tool_call = true
+structured_output = true
+knowledge = "2024-07"
+open_weights = true
+license = "MIT License"
+
+[limit]
+context = 128_000
+output = 64_000
+
+[modalities]
+input = ["text"]
+output = ["text"]
+
+[[weights]]
+label = "Hugging Face"
+url = "https://huggingface.co/deepseek-ai/DeepSeek-V3.2"

models/deepseek/deepseek-v4-pro-0813.toml

@@ -0,0 +1,19 @@
+name = "DeepSeek V4 Pro 0813"
+description = "DeepSeek V4 Pro snapshot with million-token context and support for thinking and non-thinking modes"
+family = "deepseek-thinking"
+release_date = "2026-08-12"
+last_updated = "2026-08-12"
+attachment = false
+reasoning = true
+temperature = true
+tool_call = true
+structured_output = true
+open_weights = false
+
+[limit]
+context = 1_000_000
+output = 384_000
+
+[modalities]
+input = ["text"]
+output = ["text"]

models/google/gemini-3.7-flash.toml

@@ -0,0 +1,68 @@
+name = "Gemini 3.7 Flash"
+description = "High-efficiency Gemini model for agentic workflows, coding, and multimodal reasoning"
+family = "gemini-flash"
+release_date = "2026-08-13"
+last_updated = "2026-08-13"
+attachment = true
+reasoning = true
+temperature = true
+tool_call = true
+structured_output = true
+knowledge = "2026-03"
+open_weights = false
+
+[limit]
+context = 1_048_576
+output = 65_536
+
+[modalities]
+input = ["text", "image", "video", "audio", "pdf"]
+output = ["text"]
+
+[[benchmarks]]
+name = "FrontierCode"
+score = 43.6
+metric = "score"
+version = "1.1 Main"
+source = "https://deepmind.google/models/model-cards/gemini-3-7-flash/"
+date = "2026-08-13"
+
+[[benchmarks]]
+name = "DeepSWE"
+score = 65.3
+metric = "resolve rate"
+version = "1.1"
+source = "https://deepmind.google/models/model-cards/gemini-3-7-flash/"
+date = "2026-08-13"
+
+[[benchmarks]]
+name = "Terminal-Bench"
+score = 85.8
+metric = "accuracy"
+version = "2.1"
+source = "https://deepmind.google/models/model-cards/gemini-3-7-flash/"
+date = "2026-08-13"
+
+[[benchmarks]]
+name = "AutomationBench"
+score = 30.4
+metric = "accuracy"
+dataset = "private set"
+source = "https://deepmind.google/models/model-cards/gemini-3-7-flash/"
+date = "2026-08-13"
+
+[[benchmarks]]
+name = "GDP.pdf"
+score = 34.0
+metric = "accuracy"
+source = "https://deepmind.google/models/model-cards/gemini-3-7-flash/"
+date = "2026-08-13"
+
+[[benchmarks]]
+name = "GDM-MRCR"
+score = 97.0
+metric = "accuracy"
+variant = "128k average, 8-needle"
+version = "v2"
+source = "https://deepmind.google/models/model-cards/gemini-3-7-flash/"
+date = "2026-08-13"

models/ibm/granite-4-h-micro.toml

@@ -0,0 +1,23 @@
+name = "Granite-4.0-H-Micro"
+description = "Compact open-weight hybrid Granite model for lightweight enterprise chat and tool calling"
+family = "granite"
+release_date = "2025-10-02"
+last_updated = "2025-10-02"
+attachment = false
+reasoning = false
+temperature = true
+tool_call = true
+structured_output = true
+open_weights = true
+
+[limit]
+context = 131_072
+output = 131_072
+
+[modalities]
+input = ["text"]
+output = ["text"]
+
+[[weights]]
+label = "Hugging Face"
+url = "https://huggingface.co/ibm-granite/granite-4.0-h-micro"

models/ibm/granite-4-h-small.toml

@@ -0,0 +1,23 @@
+name = "Granite-4.0-H-Small"
+description = "Open-weight hybrid model for enterprise chat, coding, retrieval-augmented generation, and tool-calling workloads"
+family = "granite"
+release_date = "2025-10-02"
+last_updated = "2025-10-02"
+attachment = false
+reasoning = false
+temperature = true
+tool_call = true
+structured_output = true
+open_weights = true
+
+[limit]
+context = 131_072
+output = 131_072
+
+[modalities]
+input = ["text"]
+output = ["text"]
+
+[[weights]]
+label = "Hugging Face"
+url = "https://huggingface.co/ibm-granite/granite-4.0-h-small"

models/meta/llama-3.1-8b-instruct.toml

@@ -0,0 +1,23 @@
+name = "Llama-3.1-8B-Instruct"
+description = "Compact open Llama model for lightweight chat, drafting, and self-hosting"
+family = "llama"
+release_date = "2024-07-23"
+last_updated = "2024-07-23"
+attachment = false
+reasoning = false
+temperature = true
+tool_call = true
+knowledge = "2023-12"
+open_weights = true
+
+[limit]
+context = 128_000
+output = 4_096
+
+[modalities]
+input = ["text"]
+output = ["text"]
+
+[[weights]]
+label = "Hugging Face"
+url = "https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct"

models/meta/llama-3.2-11b-vision-instruct.toml

@@ -0,0 +1,23 @@
+name = "Llama-3.2-11B-Vision-Instruct"
+description = "Open multimodal Llama model for image understanding, captioning, and visual QA"
+family = "llama"
+release_date = "2024-09-25"
+last_updated = "2024-09-25"
+attachment = true
+reasoning = false
+temperature = true
+tool_call = true
+knowledge = "2023-12"
+open_weights = true
+
+[limit]
+context = 128_000
+output = 4_096
+
+[modalities]
+input = ["text", "image"]
+output = ["text"]
+
+[[weights]]
+label = "Hugging Face"
+url = "https://huggingface.co/meta-llama/Llama-3.2-11B-Vision-Instruct"

models/meta/llama-guard-3-8b.toml

@@ -0,0 +1,23 @@
+name = "Llama-Guard-3-8B"
+description = "Llama 3.1-based safety classifier for moderating prompts and model responses"
+family = "llama"
+release_date = "2024-07-23"
+last_updated = "2024-07-23"
+attachment = false
+reasoning = false
+temperature = true
+tool_call = false
+knowledge = "2023-12"
+open_weights = true
+
+[limit]
+context = 128_000
+output = 4_096
+
+[modalities]
+input = ["text"]
+output = ["text"]
+
+[[weights]]
+label = "Hugging Face"
+url = "https://huggingface.co/meta-llama/Llama-Guard-3-8B"

models/meta/muse-glimmer-30b.toml

@@ -0,0 +1,111 @@
+# Sources:
+# https://research.meta.ai/blog/introducing-muse-glimmer-open-agentic-model
+# https://huggingface.co/meta-models/Muse-Glimmer-30B
+# https://developer.meta.com/ai/models/muse-glimmer/
+
+name = "Muse Glimmer 30B"
+description = "Muse Glimmer is a 30-billion-parameter open-weight multimodal model from Meta Superintelligence Labs, distilled from Muse Spark for always-on local agents, tool use, coding, and image understanding."
+family = "muse"
+release_date = "2026-08-10"
+last_updated = "2026-08-10"
+attachment = true
+reasoning = true
+temperature = true
+tool_call = true
+structured_output = true
+knowledge = "2026-01-04"
+open_weights = true
+license = "Apache 2.0"
+
+[limit]
+context = 131_072
+output = 131_072
+
+[modalities]
+input = ["text", "image"]
+output = ["text"]
+
+[[weights]]
+label = "Hugging Face"
+url = "https://huggingface.co/meta-models/Muse-Glimmer-30B"
+
+[[links]]
+label = "Announcement"
+url = "https://research.meta.ai/blog/introducing-muse-glimmer-open-agentic-model"
+type = "announcement"
+
+[[links]]
+label = "Model card"
+url = "https://huggingface.co/meta-models/Muse-Glimmer-30B"
+type = "model_card"
+
+[[links]]
+label = "Developer docs"
+url = "https://developer.meta.com/ai/models/muse-glimmer/"
+type = "docs"
+
+[[benchmarks]]
+name = "MCP Atlas"
+score = 75.5
+metric = "success rate"
+variant = "public"
+source = "https://huggingface.co/meta-models/Muse-Glimmer-30B"
+date = "2026-08-10"
+
+[[benchmarks]]
+name = "DeepSearch QA"
+score = 74.6
+source = "https://huggingface.co/meta-models/Muse-Glimmer-30B"
+date = "2026-08-10"
+
+[[benchmarks]]
+name = "SWE-Bench Pro"
+score = 51.2
+metric = "resolve rate"
+source = "https://huggingface.co/meta-models/Muse-Glimmer-30B"
+date = "2026-08-10"
+
+[[benchmarks]]
+name = "SWE-Bench Verified"
+score = 76.0
+metric = "resolve rate"
+source = "https://huggingface.co/meta-models/Muse-Glimmer-30B"
+date = "2026-08-10"
+
+[[benchmarks]]
+name = "Terminal-Bench"
+score = 51.7
+metric = "success rate"
+version = "2.1"
+variant = "with terminus2"
+source = "https://huggingface.co/meta-models/Muse-Glimmer-30B"
+date = "2026-08-10"
+
+[[benchmarks]]
+name = "OSWorld-Verified"
+score = 65.9
+metric = "success rate"
+source = "https://huggingface.co/meta-models/Muse-Glimmer-30B"
+date = "2026-08-10"
+
+[[benchmarks]]
+name = "AIME 2026"
+score = 94.7
+metric = "accuracy"
+source = "https://huggingface.co/meta-models/Muse-Glimmer-30B"
+date = "2026-08-10"
+
+[[benchmarks]]
+name = "GPQA Diamond"
+score = 83.5
+metric = "accuracy"
+variant = "AA"
+source = "https://huggingface.co/meta-models/Muse-Glimmer-30B"
+date = "2026-08-10"
+
+[[benchmarks]]
+name = "CharXiv Reasoning"
+score = 78.8
+metric = "accuracy"
+source = "https://huggingface.co/meta-models/Muse-Glimmer-30B"
+date = "2026-08-10"

models/meta/muse-spark-1.2.toml

@@ -0,0 +1,23 @@
+# Sources:
+# https://research.meta.ai/blog/introducing-muse-code-and-muse-spark-1-2
+# https://dev.meta.ai/docs/getting-started/models
+
+name = "Muse Spark 1.2"
+description = "Muse Spark 1.2 is a coding-focused update to Muse Spark 1.1 with improvements in code generation, complex debugging, codebase understanding, and end-to-end developer workflows."
+family = "muse"
+release_date = "2026-08-05"
+last_updated = "2026-08-05"
+attachment = true
+reasoning = true
+temperature = true
+tool_call = true
+structured_output = true
+open_weights = false
+
+[limit]
+context = 1_048_576
+output = 131_072
+
+[modalities]
+input = ["text", "image", "video", "pdf", "audio"]
+output = ["text"]

models/microsoft/mai-code-1.1-flash.toml

@@ -0,0 +1,23 @@
+name = "MAI-Code-1.1-Flash"
+description = "Microsoft coding model with native vision support, optimized for fast and efficient software development"
+family = "mai"
+release_date = "2026-08-11"
+last_updated = "2026-08-11"
+attachment = true
+reasoning = true
+tool_call = true
+structured_output = true
+open_weights = false
+
+[limit]
+context = 256_000
+output = 128_000
+
+[modalities]
+input = ["text", "image"]
+output = ["text"]
+
+[[links]]
+label = "Announcement"
+url = "https://microsoft.ai/news/mai-code-1-1-flash-br-better-faster-at-a-quarter-of-the-cost/"
+type = "announcement"

models/microsoft/phi-4-mini.toml

@@ -0,0 +1,30 @@
+name = "Phi-4-mini"
+description = "Compact Microsoft instruction model tuned for efficient coding assistance, reasoning, and low-latency agent tasks"
+family = "phi"
+release_date = "2024-12-11"
+last_updated = "2024-12-11"
+attachment = false
+reasoning = false
+temperature = true
+tool_call = true
+knowledge = "2023-10"
+open_weights = true
+
+[limit]
+context = 128_000
+output = 4_096
+
+[modalities]
+input = ["text"]
+output = ["text"]
+
+[[links]]
+label = "Weights"
+url = "https://huggingface.co/microsoft/Phi-4-mini-instruct"
+type = "weights"
+
+[[benchmarks]]
+name = "MMLU"
+score = 67.3
+metric = "accuracy"
+source = "https://huggingface.co/microsoft/Phi-4-mini-instruct/resolve/main/README.md"

models/minimax/MiniMax-M2-Her.toml

@@ -0,0 +1,19 @@
+# Source: https://api.ofox.ai/v2/models/catalog?include=provider_price&limit=500
+name = "MiniMax-M2 Her"
+description = "MiniMax M2 variant tuned for conversational and character-driven agent interactions"
+family = "minimax"
+release_date = "2026-01-23"
+last_updated = "2026-01-23"
+attachment = false
+reasoning = true
+temperature = true
+tool_call = true
+open_weights = false
+
+[limit]
+context = 200_000
+output = 131_000
+
+[modalities]
+input = ["text"]
+output = ["text"]

models/mistral/mistral-small-3-1-24b-instruct-2503.toml

@@ -0,0 +1,24 @@
+name = "Mistral Small 3.1 24B"
+description = "Efficient multimodal model for instruction following, coding, reasoning, and function calling"
+family = "mistral-small"
+release_date = "2025-03-17"
+last_updated = "2025-03-17"
+attachment = true
+reasoning = false
+temperature = true
+tool_call = true
+structured_output = true
+knowledge = "2024-06"
+open_weights = true
+
+[limit]
+context = 128_000
+output = 16_384
+
+[modalities]
+input = ["text", "image"]
+output = ["text"]
+
+[[weights]]
+label = "Hugging Face"
+url = "https://huggingface.co/mistralai/Mistral-Small-3.1-24B-Instruct-2503"

models/nvidia/nemotron-3.5-lightning.toml

@@ -0,0 +1,19 @@
+name = "Nemotron 3.5 Lightning 30B A3B"
+description = "Fast NVIDIA Nemotron MoE for reliable agentic tasks across enterprise workloads"
+family = "nemotron"
+release_date = "2026-08-11"
+last_updated = "2026-08-11"
+attachment = false
+reasoning = true
+temperature = true
+tool_call = true
+structured_output = true
+open_weights = true
+
+[limit]
+context = 262_144
+output = 262_144
+
+[modalities]
+input = ["text"]
+output = ["text"]

models/openai/gpt-5.3-codex-spark.toml

@@ -0,0 +1,21 @@
+name = "GPT-5.3 Codex Spark"
+description = "Coding-optimized GPT model for repository edits, reviews, and agentic software work"
+family = "gpt-codex-spark"
+release_date = "2026-02-05"
+last_updated = "2026-02-05"
+attachment = true
+reasoning = true
+temperature = false
+knowledge = "2025-08-31"
+tool_call = true
+structured_output = true
+open_weights = false
+
+[limit]
+context = 128_000
+input = 100_000
+output = 32_000
+
+[modalities]
+input = ["text", "image", "pdf"]
+output = ["text"]

models/swiss-ai/apertus-70b.toml

@@ -0,0 +1,28 @@
+name = "Apertus 70B"
+description = "Fully open 70B multilingual LLM supporting 1800+ languages with 65K context. Trained on 15T tokens of compliant open data. Apache 2.0, EU AI Act compliant."
+release_date = "2025-09-02"
+last_updated = "2025-09-02"
+knowledge = "2025-09"
+attachment = false
+reasoning = false
+temperature = true
+tool_call = true
+open_weights = true
+license = "Apache-2.0"
+
+[limit]
+context = 65_536
+output = 8_192
+
+[modalities]
+input = ["text"]
+output = ["text"]
+
+[[weights]]
+label = "Hugging Face"
+url = "https://huggingface.co/swiss-ai/Apertus-70B-Instruct-2509"
+
+[[links]]
+label = "Paper"
+url = "https://arxiv.org/abs/2509.14233"
+type = "paper"

models/xai/grok-4.1-fast.toml

@@ -0,0 +1,24 @@
+# xAI Grok 4.1 Fast (non-reasoning).
+# Sources:
+# - https://x.ai/news/grok-4-1-fast (release 2025-11-19; variants + $0.20/$0.50/$0.05 pricing)
+# - https://docs.oracle.com/en-us/iaas/Content/generative-ai/xai-grok-4-1-fast.htm (2M context, text+image, tools, structured outputs, non-reasoning mode)
+# - https://api.ofox.ai/v1/models/x-ai/grok-4.1-fast (canonical_slug grok-4-1-fast-non-reasoning; context 2M; max_completion 30k)
+name = "Grok 4.1 Fast"
+description = "xAI's fast agentic tool-calling model with a 2M context window; non-reasoning variant for low-latency responses"
+family = "grok"
+release_date = "2025-11-19"
+last_updated = "2025-11-19"
+attachment = true
+reasoning = false
+temperature = true
+tool_call = true
+structured_output = true
+open_weights = false
+
+[limit]
+context = 2_000_000
+output = 30_000
+
+[modalities]
+input = ["text", "image"]
+output = ["text"]

models/xai/grok-4.5.toml

@@ -1,5 +1,5 @@
 name = "Grok 4.5"
-description = "xAI's latest Grok for chat, coding, agentic tools, and lower hallucination risk"
+description = "xAI's Grok model for chat, coding, agentic tools, and lower hallucination risk"
 family = "grok"
 release_date = "2026-07-08"
 last_updated = "2026-07-08"

models/xai/grok-4.6.toml

@@ -0,0 +1,20 @@
+name = "Grok 4.6"
+description = "xAI's frontier model for long-running agents, coding, knowledge work, and visual projects"
+family = "grok"
+knowledge = "2026-02-01"
+release_date = "2026-08-12"
+last_updated = "2026-08-12"
+attachment = true
+reasoning = true
+temperature = true
+tool_call = true
+structured_output = true
+open_weights = false
+
+[limit]
+context = 500_000
+output = 500_000
+
+[modalities]
+input = ["text", "image"]
+output = ["text"]

models/zhipuai/glm-5.3.toml

@@ -0,0 +1,19 @@
+name = "GLM-5.3"
+description = "Flagship GLM model for long-horizon coding, agents, and complex project delivery"
+family = "glm"
+release_date = "2026-08-14"
+last_updated = "2026-08-14"
+attachment = false
+reasoning = true
+temperature = true
+tool_call = true
+structured_output = true
+open_weights = false
+
+[limit]
+context = 1_000_000
+output = 131_072
+
+[modalities]
+input = ["text"]
+output = ["text"]

packages/core/src/family.ts

@@ -30,6 +30,7 @@ export const ModelFamilyValues = [
   "claude-sonnet",
   "claude-opus",
   "claude-fable",
+  "claude-mythos",
 
   // Gemini style
   "gemini",

packages/core/src/schema.ts

@@ -397,6 +397,7 @@ export const Provider = z
       const isOpenAI = data.npm === "@ai-sdk/openai";
       const isOpenAIcompatible = data.npm === "@ai-sdk/openai-compatible";
       const isOpenrouter = data.npm === "@openrouter/ai-sdk-provider";
+      const isMergeGateway = data.npm === "merge-gateway-ai-sdk-provider";
       const isAnthropic = data.npm === "@ai-sdk/anthropic";
       const isKiro = data.npm === "kiro-acp-ai-provider";
       const hasApi = data.api !== undefined;
@@ -406,6 +407,8 @@ export const Provider = z
         (isOpenAIcompatible && hasApi) ||
         // openrouter: must have api
         (isOpenrouter && hasApi) ||
+        // Merge Gateway: native provider with an OpenAI-compatible fallback
+        (isMergeGateway && hasApi) ||
         // anthropic: api optional (always allowed)
         isAnthropic ||
         // openai: api optional (always allowed)
@@ -416,6 +419,7 @@ export const Provider = z
         (!isOpenAI &&
           !isOpenAIcompatible &&
           !isOpenrouter &&
+          !isMergeGateway &&
           !isAnthropic &&
           !isKiro &&
           !hasApi)
@@ -423,7 +427,7 @@ export const Provider = z
     },
     {
       message:
-        "'api' is required for openai-compatible and openrouter, optional for anthropic, openai, and kiro, forbidden otherwise",
+        "'api' is required for openai-compatible, openrouter, and Merge Gateway; optional for anthropic, openai, and kiro; forbidden otherwise",
       path: ["api"],
     },
   );

packages/core/src/sync/auto-merge.ts

@@ -4,7 +4,15 @@ import { isDeepStrictEqual } from "node:util";
 export const MAX_CREATED_MODELS = 10;
 export const MAX_DELETED_MODELS = 10;
 export const MAX_MODEL_CHURN = 15;
-const REVIEWED_REASONING_PROVIDERS = new Set(["openrouter"]);
+const REVIEWED_REASONING_PROVIDERS = new Set([
+  "empiriolabs",
+  "kilo",
+  "llmgateway",
+  "merge-gateway",
+  "nano-gpt",
+  "openrouter",
+  "venice",
+]);
 
 export interface CatalogChange {
   status: "created" | "updated" | "deleted";

packages/core/src/sync/index.ts

@@ -14,10 +14,12 @@ import { cortecs } from "./providers/cortecs.js";
 import { crossmodel } from "./providers/crossmodel.js";
 import { deepinfra } from "./providers/deepinfra.js";
 import { digitalocean } from "./providers/digitalocean.js";
+import { edenai } from "./providers/edenai.js";
 import { empiriolabs } from "./providers/empiriolabs.js";
 import { google } from "./providers/google.js";
 import { hyper } from "./providers/hyper.js";
 import { huggingface } from "./providers/huggingface.js";
+import { inceptron } from "./providers/inceptron.js";
 import { kilo } from "./providers/kilo.js";
 import { llmgateway } from "./providers/llmgateway.js";
 import { mergeGateway } from "./providers/merge-gateway.js";
@@ -121,10 +123,12 @@ export const providers: {
   crossmodel: SyncProvider<any>;
   deepinfra: SyncProvider<any>;
   digitalocean: SyncProvider<any>;
+  edenai: SyncProvider<any>;
   empiriolabs: SyncProvider<any>;
   google: SyncProvider<any>;
   hyper: SyncProvider<any>;
   huggingface: SyncProvider<any>;
+  inceptron: SyncProvider<any>;
   kilo: SyncProvider<any>;
   llmgateway: SyncProvider<any>;
   "merge-gateway": SyncProvider<any>;
@@ -150,10 +154,12 @@ export const providers: {
   crossmodel,
   deepinfra,
   digitalocean,
+  edenai,
   empiriolabs,
   google,
   hyper,
   huggingface,
+  inceptron,
   kilo,
   llmgateway,
   "merge-gateway": mergeGateway,
@@ -174,8 +180,10 @@ export const providers: {
 export const groups = {
   aggregators: [
     "crossmodel",
+    "edenai",
     "empiriolabs",
     "huggingface",
+    "inceptron",
     "kilo",
     "llmgateway",
     "merge-gateway",

packages/core/src/sync/providers/cortecs.ts

@@ -101,7 +101,7 @@ export function buildCortecsModel(
     return factorBaseModel(canonical, {
       description: existing?.description,
       attachment: input.some((value) => value !== "text"),
-      reasoning: undefined,
+      reasoning,
       reasoning_options: reasoningOptions,
       temperature: existing?.temperature,
       tool_call: features.has("tools"),

packages/core/src/sync/providers/crossmodel.ts

@@ -24,11 +24,13 @@ const API_ENDPOINT = process.env.CROSSMODEL_MODELS_URL ?? "https://www.crossmode
 
 const ReasoningCapability = z
   .object({
-    toggle: z.boolean().optional(),
-    effort: z.array(z.string()).optional(),
+    supported: z.boolean().optional(),
+    toggle: z.boolean().nullish().transform((value) => value ?? undefined),
+    effort: z.array(z.string()).nullish().transform((value) => value ?? undefined),
     budget_tokens: z
       .object({ min: z.number().optional(), max: z.number().optional() })
-      .optional(),
+      .nullish()
+      .transform((value) => value ?? undefined),
   })
   .passthrough();
 
@@ -173,7 +175,7 @@ function isReasoningEffort(value: string): value is ReasoningEffort {
 //   otherwise         -> toggle / effort / budget_tokens entries
 function reasoningOptions(model: CrossModelModel): SyncedModel["reasoning_options"] {
   const reasoning = model.capabilities?.reasoning;
-  if (reasoning === undefined) return undefined;
+  if (reasoning === undefined || reasoning.supported === false) return undefined;
   const options: NonNullable<SyncedModel["reasoning_options"]> = [];
   if (reasoning.toggle === true) options.push({ type: "toggle" });
   if (reasoning.effort !== undefined) {

packages/core/src/sync/providers/deepinfra.ts

@@ -372,6 +372,7 @@ export function buildDeepInfraModel(
 // catalog's canonical metadata namespace so new models can inherit via
 // `base_model` whenever a `models/` entry already exists.
 const DEEPINFRA_PREFIXES: Record<string, string> = {
+  ByteDance: "bytedance-seed",
   "deepseek-ai": "deepseek",
   "meta-llama": "meta",
   google: "google",

packages/core/src/sync/providers/edenai.ts

@@ -0,0 +1,506 @@
+import { existsSync, readdirSync, readFileSync } from "node:fs";
+import path from "node:path";
+
+import { z } from "zod";
+
+import type { SyncProvider, SyncedFullModel, SyncedModel } from "../index.js";
+import {
+  factorBaseModel,
+  modelMetadata,
+  resolveModelMetadataBaseModel,
+} from "./openrouter.js";
+
+// ========================================
+// Constants
+// ========================================
+
+const API_ENDPOINT = "https://api.edenai.run/v3/models";
+const MODELS_DIR = path.join(
+  import.meta.dirname,
+  "..",
+  "..",
+  "..",
+  "..",
+  "..",
+  "models",
+);
+const PROVIDERS_DIR = path.join(MODELS_DIR, "..", "providers");
+const TOKENS_PER_MILLION = 1_000_000;
+const PRICE_DECIMALS = 1_000_000;
+
+// Values `reasoning_effort` accepts on POST /v3/chat/completions. Which of them
+// a given model exposes comes from its lab entry, not from this list.
+const ACCEPTED_EFFORTS = new Set([
+  "none",
+  "minimal",
+  "low",
+  "medium",
+  "high",
+  "xhigh",
+  "max",
+]);
+
+const REGION_SUFFIX = /@[a-z0-9-]+$/i;
+const DOTTED_VENDOR = /^[a-z0-9-]+\./;
+const VERSION_TAIL = /-v\d+:\d+$/;
+const DATE_TAIL = /-\d{8}$/;
+const DATABRICKS_PREFIX = "databricks-";
+const TIER_KEY = /^input_cost_per_token_above_(\d+)k_tokens$/;
+
+const MODALITY_BY_EDENAI: Record<
+  string,
+  SyncedFullModel["modalities"]["input"][number]
+> = {
+  text: "text",
+  image: "image",
+  audio: "audio",
+  video: "video",
+  file: "pdf",
+};
+
+// Upstreams that are the lab's own API for models under that namespace.
+const LAB_UPSTREAMS: Record<string, readonly string[]> = {
+  alibaba: ["qwen"],
+  amazon: ["amazon"],
+  anthropic: ["anthropic"],
+  cohere: ["cohere"],
+  deepseek: ["deepseek"],
+  google: ["google", "vertex"],
+  microsoft: ["microsoft"],
+  minimax: ["minimax"],
+  mistral: ["mistral"],
+  moonshotai: ["moonshot"],
+  openai: ["openai"],
+  perplexity: ["perplexityai"],
+  xai: ["xai"],
+  zhipuai: ["zai"],
+};
+
+type ReasoningOption = NonNullable<
+  SyncedFullModel["reasoning_options"]
+>[number];
+
+const canonicalNameByID = new Map<string, string>();
+
+let firstPartyBaseModels: ReadonlySet<string> = new Set();
+
+// ========================================
+// Schemas
+// ========================================
+
+const EdenAIPricing = z
+  .object({
+    input_cost_per_token: z.number().nullish(),
+    output_cost_per_token: z.number().nullish(),
+    output_cost_per_reasoning_token: z.number().nullish(),
+    cache_read_input_token_cost: z.number().nullish(),
+    cache_creation_input_token_cost: z.number().nullish(),
+    input_cost_per_audio_token: z.number().nullish(),
+  })
+  .passthrough();
+
+const EdenAICapabilities = z
+  .object({
+    input_modalities: z.array(z.string()).nullish(),
+    output_modalities: z.array(z.string()).nullish(),
+    supports_function_calling: z.boolean().optional(),
+    supports_response_schema: z.boolean().optional(),
+  })
+  .passthrough();
+
+export const EdenAIModel = z
+  .object({
+    id: z.string().min(1),
+    owned_by: z.string().min(1),
+    model_name: z.string().min(1),
+    context_length: z.number().nullish(),
+    capabilities: EdenAICapabilities,
+    pricing: EdenAIPricing.nullish(),
+    list_pricing: EdenAIPricing.nullish(),
+    alias_of: z.string().nullish(),
+  })
+  .passthrough();
+
+export const EdenAIResponse = z
+  .object({
+    object: z.literal("list"),
+    data: z.array(EdenAIModel),
+  })
+  .passthrough();
+
+export type EdenAIModel = z.infer<typeof EdenAIModel>;
+
+// ========================================
+// Base model resolution
+// ========================================
+
+function baseModelExists(modelID: string) {
+  return existsSync(path.join(MODELS_DIR, `${modelID}.toml`));
+}
+
+// Ids are `<upstream>/<native id>`, so each upstream keeps its own convention.
+function baseModelCandidates(modelName: string) {
+  const candidates = [modelName];
+
+  if (DOTTED_VENDOR.test(modelName)) {
+    const dotted = modelName.replace(".", "/");
+    candidates.push(
+      dotted,
+      dotted.replace(VERSION_TAIL, "").replace(DATE_TAIL, ""),
+    );
+  }
+
+  const last = modelName.split("/").at(-1) ?? modelName;
+  candidates.push(last);
+  if (last.startsWith(DATABRICKS_PREFIX)) {
+    candidates.push(last.slice(DATABRICKS_PREFIX.length));
+  }
+  candidates.push(last.replace(VERSION_TAIL, "").replace(DATE_TAIL, ""));
+
+  return [...new Set(candidates)].filter((candidate) => candidate.length > 0);
+}
+
+export function resolveEdenAIBaseModel(model: EdenAIModel) {
+  const names = [model.model_name.replace(REGION_SUFFIX, "")];
+  if (model.alias_of != null) {
+    const target = model.alias_of.split("/").slice(1).join("/");
+    if (target.length > 0) names.push(target);
+  }
+
+  for (const name of names) {
+    for (const candidate of baseModelCandidates(name)) {
+      const resolved = resolveModelMetadataBaseModel(candidate);
+      if (resolved !== undefined && baseModelExists(resolved)) return resolved;
+    }
+  }
+  return undefined;
+}
+
+function isFirstPartyRoute(model: EdenAIModel, baseModel: string) {
+  const lab = baseModel.split("/")[0] ?? "";
+  return (LAB_UPSTREAMS[lab] ?? []).includes(model.owned_by);
+}
+
+export function collectFirstPartyBaseModels(models: readonly EdenAIModel[]) {
+  const bases = new Set<string>();
+  for (const model of models) {
+    const baseModel = resolveEdenAIBaseModel(model);
+    if (baseModel !== undefined && isFirstPartyRoute(model, baseModel)) {
+      bases.add(baseModel);
+    }
+  }
+  return bases;
+}
+
+function canonicalModelName(baseModel: string) {
+  let cached = canonicalNameByID.get(baseModel);
+  if (cached === undefined) {
+    try {
+      const toml = Bun.TOML.parse(
+        readFileSync(path.join(MODELS_DIR, `${baseModel}.toml`), "utf8"),
+      ) as { name?: unknown };
+      cached = typeof toml.name === "string" ? toml.name : "";
+    } catch {
+      cached = "";
+    }
+    canonicalNameByID.set(baseModel, cached);
+  }
+  return cached === "" ? undefined : cached;
+}
+
+function hasOutputLimit(baseModel: string) {
+  const limit = modelMetadata(baseModel).limit;
+  return (
+    typeof limit === "object" &&
+    limit !== null &&
+    typeof (limit as { output?: unknown }).output === "number"
+  );
+}
+
+function regionVariantName(model: EdenAIModel, baseModel: string) {
+  const region = REGION_SUFFIX.exec(model.id)?.[0].slice(1);
+  if (region === undefined) return undefined;
+
+  const canonical = canonicalModelName(baseModel);
+  if (canonical === undefined) return undefined;
+  return `${canonical} (${region.toUpperCase()})`;
+}
+
+// ========================================
+// Reasoning options
+// ========================================
+
+// Eden AI's only reasoning control is `reasoning_effort`, so a model is
+// published with the effort list its lab entry (or an established relay peer)
+// already documents. Peers exposing only `toggle` / `budget_tokens` have no
+// equivalent here, and those models are skipped rather than given a guess.
+function effortValues(options: unknown): string[] | "always-on" | undefined {
+  if (!Array.isArray(options)) return undefined;
+  if (options.length === 0) return "always-on";
+
+  let toggled = false;
+  let accepted: string[] | undefined;
+
+  for (const option of options) {
+    if (typeof option !== "object" || option === null) continue;
+    const type = (option as { type?: unknown }).type;
+    if (type === "toggle") toggled = true;
+    if (type !== "effort") continue;
+
+    const values = (option as { values?: unknown }).values;
+    if (!Array.isArray(values)) continue;
+    const filtered = values.filter(
+      (value): value is string =>
+        typeof value === "string" && ACCEPTED_EFFORTS.has(value),
+    );
+    if (filtered.length > 0) accepted = filtered;
+  }
+
+  if (accepted === undefined) return undefined;
+  // Eden AI switches reasoning off with `reasoning_effort=none`, so a lab-side
+  // toggle becomes `none` in the effort list instead of a separate option.
+  return toggled && !accepted.includes("none")
+    ? ["none", ...accepted]
+    : accepted;
+}
+
+function parseToml(filePath: string) {
+  try {
+    return Bun.TOML.parse(readFileSync(filePath, "utf8")) as Record<
+      string,
+      unknown
+    >;
+  } catch {
+    return undefined;
+  }
+}
+
+function tomlFilesIn(dir: string): string[] {
+  let entries;
+  try {
+    entries = readdirSync(dir, { withFileTypes: true });
+  } catch {
+    return [];
+  }
+  return entries.flatMap((entry) =>
+    entry.isDirectory()
+      ? tomlFilesIn(path.join(dir, entry.name))
+      : entry.name.endsWith(".toml")
+        ? [path.join(dir, entry.name)]
+        : [],
+  );
+}
+
+// OpenRouter is the established same-surface relay, so it is the one peer
+// consulted when a lab entry documents no effort levels.
+const PEER_PROVIDER = "openrouter";
+
+let peerEfforts: Map<string, string[] | "always-on"> | undefined;
+
+function peerEffortsByBaseModel() {
+  if (peerEfforts !== undefined) return peerEfforts;
+
+  peerEfforts = new Map();
+  for (const file of tomlFilesIn(
+    path.join(PROVIDERS_DIR, PEER_PROVIDER, "models"),
+  )) {
+    const toml = parseToml(file);
+    const base = toml?.base_model;
+    if (typeof base !== "string" || peerEfforts.has(base)) continue;
+    const values = effortValues(toml?.reasoning_options);
+    if (values !== undefined) peerEfforts.set(base, values);
+  }
+  return peerEfforts;
+}
+
+export function reasoningOptionsFor(
+  baseModel: string,
+): SyncedFullModel["reasoning_options"] | undefined {
+  const [lab, ...rest] = baseModel.split("/");
+  const firstParty = effortValues(
+    parseToml(
+      path.join(PROVIDERS_DIR, lab ?? "", "models", `${rest.join("/")}.toml`),
+    )?.reasoning_options,
+  );
+  const derived = firstParty ?? peerEffortsByBaseModel().get(baseModel);
+
+  if (derived === undefined) return undefined;
+  if (derived === "always-on") return [];
+  return [{ type: "effort", values: derived } as ReasoningOption];
+}
+
+// ========================================
+// Cost
+// ========================================
+
+function pricePerMillion(price: number) {
+  return (
+    Math.round(price * TOKENS_PER_MILLION * PRICE_DECIMALS) / PRICE_DECIMALS
+  );
+}
+
+function chargedPricePerMillion(price: unknown) {
+  return typeof price === "number" && price > 0
+    ? pricePerMillion(price)
+    : undefined;
+}
+
+function costTiers(pricing: Record<string, unknown>) {
+  const thresholds = Object.keys(pricing)
+    .map((key) => TIER_KEY.exec(key)?.[1])
+    .filter((value): value is string => value !== undefined)
+    .map(Number)
+    .sort((a, b) => a - b);
+
+  return thresholds.flatMap((threshold) => {
+    // Built explicitly so `..._above_1hr_above_200k_tokens` is never read as a
+    // context tier.
+    const suffix = `_above_${threshold}k_tokens`;
+    const input = pricing[`input_cost_per_token${suffix}`];
+    const output = pricing[`output_cost_per_token${suffix}`];
+    if (typeof input !== "number" || typeof output !== "number") return [];
+
+    return [
+      {
+        tier: { type: "context" as const, size: threshold * 1_000 },
+        input: pricePerMillion(input),
+        output: pricePerMillion(output),
+        cache_read: chargedPricePerMillion(
+          pricing[`cache_read_input_token_cost${suffix}`],
+        ),
+        cache_write: chargedPricePerMillion(
+          pricing[`cache_creation_input_token_cost${suffix}`],
+        ),
+      },
+    ];
+  });
+}
+
+function buildCost(
+  model: EdenAIModel,
+  reasoning: boolean,
+): SyncedFullModel["cost"] {
+  // `pricing` carries account-level discounts; `list_pricing` is the public rate.
+  const pricing = model.list_pricing ?? model.pricing;
+  if (pricing == null) return undefined;
+
+  const input = pricing.input_cost_per_token;
+  const output = pricing.output_cost_per_token;
+  if (input == null || output == null) return undefined;
+
+  const tiers = costTiers(pricing);
+  return {
+    input: pricePerMillion(input),
+    output: pricePerMillion(output),
+    reasoning: reasoning
+      ? chargedPricePerMillion(pricing.output_cost_per_reasoning_token)
+      : undefined,
+    cache_read: chargedPricePerMillion(pricing.cache_read_input_token_cost),
+    cache_write: chargedPricePerMillion(
+      pricing.cache_creation_input_token_cost,
+    ),
+    input_audio: chargedPricePerMillion(pricing.input_cost_per_audio_token),
+    tiers: tiers.length > 0 ? tiers : undefined,
+  };
+}
+
+// ========================================
+// Model translation
+// ========================================
+
+function mapModalities(values: readonly string[] | null | undefined) {
+  if (values == null) return undefined;
+
+  const mapped = [
+    ...new Set(
+      values
+        .map((value) => MODALITY_BY_EDENAI[value.toLowerCase()])
+        .filter(
+          (value): value is NonNullable<typeof value> => value !== undefined,
+        ),
+    ),
+  ];
+  return mapped.length > 0 ? mapped : undefined;
+}
+
+export function buildEdenAIModel(
+  model: EdenAIModel,
+  firstParty: ReadonlySet<string> = firstPartyBaseModels,
+): SyncedModel | undefined {
+  const baseModel = resolveEdenAIBaseModel(model);
+  // Eden AI relays other labs' models only, so an entry needs its lab metadata.
+  if (baseModel === undefined) return undefined;
+  // The catalog reports no output limit, so the base has to resolve one.
+  if (!hasOutputLimit(baseModel)) return undefined;
+  // Where Eden AI relays the lab's own API, that route is the entry. Models
+  // with no first-party route keep every route, since their prices differ and
+  // there is no canonical one to pick.
+  if (firstParty.has(baseModel) && !isFirstPartyRoute(model, baseModel)) {
+    return undefined;
+  }
+
+  const capabilities = model.capabilities;
+  const input = mapModalities(capabilities.input_modalities);
+  const output = mapModalities(capabilities.output_modalities);
+  const modalities =
+    input !== undefined && output !== undefined ? { input, output } : undefined;
+
+  // Whether a model reasons is a property of the model, not of the relay, so
+  // the lab entry owns it and only the effort controls are authored here.
+  const reasoning = modelMetadata(baseModel).reasoning === true;
+  const reasoningOptions = reasoning
+    ? reasoningOptionsFor(baseModel)
+    : undefined;
+  if (reasoning && reasoningOptions === undefined) return undefined;
+
+  const limit =
+    model.context_length != null && model.context_length > 0
+      ? { context: model.context_length }
+      : undefined;
+
+  return factorBaseModel(
+    baseModel,
+    {
+      name: regionVariantName(model, baseModel),
+      modalities,
+      attachment: input?.some((value) => value !== "text"),
+      reasoning_options: reasoningOptions,
+      tool_call: capabilities.supports_function_calling,
+      structured_output: capabilities.supports_response_schema,
+      cost: buildCost(model, reasoning),
+      limit,
+    },
+    limit,
+  );
+}
+
+// ========================================
+// Eden AI provider
+// ========================================
+
+export const edenai = {
+  id: "edenai",
+  name: "Eden AI",
+  modelsDir: "providers/edenai/models",
+  preserveBaseModels: false,
+  preserveDescriptions: false,
+  async fetchModels() {
+    const response = await fetch(API_ENDPOINT);
+    if (!response.ok) {
+      throw new Error(
+        `Eden AI request failed: ${response.status} ${response.statusText}`,
+      );
+    }
+    return response.json();
+  },
+  parseModels(raw) {
+    const models = EdenAIResponse.parse(raw).data;
+    firstPartyBaseModels = collectFirstPartyBaseModels(models);
+    return models;
+  },
+  translateModel(model) {
+    const built = buildEdenAIModel(model);
+    if (built === undefined) return undefined;
+    return { id: model.id, model: built };
+  },
+} satisfies SyncProvider<EdenAIModel>;

packages/core/src/sync/providers/empiriolabs.ts

@@ -1,26 +1,18 @@
 import { z } from "zod";
 
 import type { ExistingModel, SyncProvider, SyncedFullModel, SyncedModel } from "../index.js";
-import { factorBaseModel, resolveCanonicalBaseModel } from "./openrouter.js";
+import { factorBaseModel, resolveCanonicalBaseModel, resolveModelMetadataBaseModel } from "./openrouter.js";
 
 // EmpirioLabs exposes a public, unauthenticated OpenAI-compatible model
 // catalog, so no API key is needed or used for this sync.
 const API_ENDPOINT = "https://api.empiriolabs.ai/v1/models";
 
+// Keep this for slugs that cannot be derived from a lab filename.
+// Family prefixes, version-dot slugs, unique filenames, and dated/version
+// suffixes are resolved automatically by resolveEmpiriolabsBaseModel.
 const CANONICAL_BASE_MODELS: Record<string, string> = {
-  "fugu-ultra": "sakana/fugu-ultra",
-  "deepseek-v4-flash-0731": "deepseek/deepseek-v4-flash-0731",
-  "gemma-4-26b-a4b": "google/gemma-4-26b-a4b-it",
-  "gemma-4-e4b": "google/gemma-4-E4B-it",
   "mistral-medium-3": "mistral/mistral-medium-2505",
   "mistral-small-4": "mistral/mistral-small-2603",
-  "muse-spark-1-1": "meta/muse-spark-1.1",
-  "qwen3-5-9b": "alibaba/qwen3.5-9b",
-  "qwen3-7-max": "alibaba/qwen3.7-max",
-  "qwen3-7-plus": "alibaba/qwen3.7-plus",
-  "step-3-5-flash": "stepfun/step-3.5-flash",
-  "step-3-5-flash-2603": "stepfun/step-3.5-flash-2603",
-  "step-3-7-flash": "stepfun/step-3.7-flash",
 };
 
 const EmpiriolabsParameter = z
@@ -209,55 +201,75 @@ function parameterOutputLimit(model: EmpiriolabsModel) {
   return parameter?.max !== undefined && parameter.max > 0 ? parameter.max : undefined;
 }
 
+function applyVersionDots(id: string) {
+  return id
+    .replace(/^(qwen\d+)-(\d+)/, "$1.$2")
+    .replace(/^(seed-\d+)-(\d+)/, "$1.$2")
+    .replace(/^(muse-[a-z]+)-(\d+)-(\d+)$/, "$1-$2.$3")
+    .replace(/^(glm-\d+)-(\d+)/, "$1.$2")
+    .replace(/^(kimi-k\d+)-(\d+)/, "$1.$2")
+    .replace(/^(minimax-m\d+)-(\d+)/, "$1.$2")
+    .replace(/^(mimo-v\d+)-(\d+)/, "$1.$2")
+    .replace(/^(deepseek-v\d+)-(\d+)/, "$1.$2")
+    .replace(/^(step-\d+)-(\d+)/, "$1.$2");
+}
+
+function stripProductSuffixes(id: string) {
+  const out: string[] = [];
+  if (/-v\d+(-\d+)?$/.test(id)) {
+    const dropPatch = id.replace(/-\d+$/, "");
+    if (dropPatch !== id) out.push(dropPatch);
+    out.push(id.replace(/-v\d+(-\d+)?$/, ""));
+  }
+  if (/-\d{4}$/.test(id)) out.push(id.replace(/-\d{4}$/, ""));
+  return out;
+}
+
+function idVariants(id: string) {
+  const variants = [id];
+  const dotted = applyVersionDots(id);
+  if (dotted !== id) variants.push(dotted);
+  for (const stripped of stripProductSuffixes(id)) {
+    if (!variants.includes(stripped)) variants.push(stripped);
+    const strippedDotted = applyVersionDots(stripped);
+    if (!variants.includes(strippedDotted)) variants.push(strippedDotted);
+  }
+  return variants;
+}
+
+function prefixesFor(id: string) {
+  if (id.startsWith("deepseek-")) return ["deepseek"];
+  if (id.startsWith("glm-")) return ["z-ai"];
+  if (id.startsWith("kimi-")) return ["moonshotai"];
+  if (id.startsWith("minimax-")) return ["minimax"];
+  if (id.startsWith("mimo-")) return ["xiaomi"];
+  if (id.startsWith("qwen")) return ["qwen"];
+  if (id.startsWith("muse-")) return ["meta"];
+  if (id.startsWith("seed-")) return ["bytedance-seed"];
+  if (id.startsWith("fugu-")) return ["sakana"];
+  if (id.startsWith("gemma-")) return ["google"];
+  if (id.startsWith("step") && !id.startsWith("stepaudio")) return ["stepfun"];
+  if (id.startsWith("mistral-")) return ["mistralai"];
+  return [];
+}
+
 export function resolveEmpiriolabsBaseModel(id: string) {
   const explicit = CANONICAL_BASE_MODELS[id];
   if (explicit !== undefined) return explicit;
-  return canonicalCandidates(id)
-    .map((candidate) => resolveCanonicalBaseModel(candidate))
-    .find((candidate) => candidate !== undefined);
-}
 
-function canonicalCandidates(id: string) {
-  const candidates: string[] = [];
-
-  if (id.startsWith("deepseek-")) {
-    candidates.push(`deepseek/${id}`);
-    candidates.push(`deepseek/${id.replace(/^deepseek-v(\d+)-(\d+)/, "deepseek-v$1.$2")}`);
+  for (const variant of idVariants(id)) {
+    for (const prefix of prefixesFor(variant)) {
+      const resolved = resolveCanonicalBaseModel(`${prefix}/${variant}`);
+      if (resolved !== undefined) return resolved;
+      if (prefix === "google" && !variant.endsWith("-it")) {
+        const instruct = resolveCanonicalBaseModel(`${prefix}/${variant}-it`);
+        if (instruct !== undefined) return instruct;
+      }
+    }
+    const unique = resolveModelMetadataBaseModel(variant);
+    if (unique !== undefined) return unique;
   }
-
-  if (id.startsWith("glm-")) {
-    const normalized = id
-      .replace(/^glm-(\d+)-(\d+)/, "glm-$1.$2")
-      .replace(/^glm-(\d+)-(\d+)v/, "glm-$1.$2v");
-    candidates.push(`z-ai/${id}`);
-    candidates.push(`z-ai/${normalized}`);
-  }
-
-  if (id.startsWith("kimi-")) {
-    const normalized = id.replace(/^(kimi-k\d+)-(\d+)/, "$1.$2");
-    candidates.push(`moonshotai/${id}`);
-    candidates.push(`moonshotai/${normalized}`);
-  }
-
-  if (id.startsWith("minimax-")) {
-    const normalized = id.replace(/^minimax-m(\d+)-(\d+)/, "minimax-m$1.$2");
-    candidates.push(`minimax/${id}`);
-    candidates.push(`minimax/${normalized}`);
-  }
-
-  if (id.startsWith("mimo-")) {
-    const normalized = id.replace(/^mimo-v(\d+)-(\d+)/, "mimo-v$1.$2");
-    candidates.push(`xiaomi/${id}`);
-    candidates.push(`xiaomi/${normalized}`);
-  }
-
-  if (id.startsWith("qwen")) {
-    const normalized = id.replace(/^(qwen\d+)-(\d+)/, "$1.$2");
-    candidates.push(`qwen/${id}`);
-    candidates.push(`qwen/${normalized}`);
-  }
-
-  return [...new Set(candidates)];
+  return undefined;
 }
 
 export function buildEmpiriolabsModel(

packages/core/src/sync/providers/inceptron.ts

@@ -0,0 +1,208 @@
+import { existsSync } from "node:fs";
+import path from "node:path";
+
+import { z } from "zod";
+
+import { ReasoningOption } from "../../schema.js";
+import type { SyncProvider, SyncedModel } from "../index.js";
+import { factorBaseModel } from "./openrouter.js";
+
+const API_ENDPOINT = process.env.INCEPTRON_MODELS_URL ?? "https://api.inceptron.io/v1/models";
+const MODELS_DIR = path.join(import.meta.dirname, "..", "..", "..", "..", "..", "models");
+
+const ModelsDevMetadata = z
+  .object({
+    base_model: z.string().regex(/^[^./\\][^/\\]*\/[^./\\][^/\\]*$/),
+    reasoning_options: z.array(ReasoningOption).optional(),
+    interleaved: z
+      .union([
+        z.literal(true),
+        z.object({ field: z.enum(["reasoning_content", "reasoning_details"]) }).strict(),
+      ])
+      .optional(),
+    status: z.enum(["alpha", "beta", "deprecated"]).optional(),
+  })
+  .strict();
+
+export const InceptronModel = z.object({
+  id: z.string().min(1),
+  name: z.string().min(1),
+  is_ready: z.boolean().optional(),
+  context_length: z.number().int().positive(),
+  max_output_length: z.number().int().positive(),
+  input_modalities: z.array(z.string()).min(1),
+  output_modalities: z.array(z.string()).min(1),
+  supported_features: z.array(z.string()),
+  supported_sampling_parameters: z.array(z.string()),
+  pricing: z.object({
+    prompt: z.string(),
+    completion: z.string(),
+    input_cache_reads: z.string().optional(),
+    input_cache_writes: z.string().optional(),
+  }),
+  models_dev: ModelsDevMetadata.optional(),
+});
+
+export const InceptronResponse = z
+  .object({
+    object: z.literal("list"),
+    data: z.array(InceptronModel),
+  })
+  .strict();
+
+export type InceptronModel = z.infer<typeof InceptronModel>;
+export type ReadyInceptronModel = InceptronModel & {
+  models_dev: z.infer<typeof ModelsDevMetadata>;
+};
+
+export const inceptron = {
+  id: "inceptron",
+  name: "Inceptron",
+  modelsDir: "providers/inceptron/models",
+  async fetchModels() {
+    const response = await fetch(API_ENDPOINT);
+    if (!response.ok) {
+      throw new Error(`Inceptron request failed: ${response.status} ${response.statusText}`);
+    }
+    return response.json();
+  },
+  parseModels: parseInceptronModels,
+  translateModel(model) {
+    return { id: model.id, model: buildInceptronModel(model) };
+  },
+} satisfies SyncProvider<ReadyInceptronModel>;
+
+export function parseInceptronModels(raw: unknown): ReadyInceptronModel[] {
+  const models = InceptronResponse.parse(raw).data.filter((model) => model.is_ready !== false);
+  const seen = new Set<string>();
+
+  return models.map((model) => {
+    if (seen.has(model.id)) throw new Error(`Duplicate ready Inceptron model ID: ${model.id}`);
+    seen.add(model.id);
+    if (model.models_dev === undefined) {
+      throw new Error(`Ready Inceptron model ${model.id} is missing models_dev metadata`);
+    }
+    if (!baseModelExists(model.models_dev.base_model)) {
+      throw new Error(
+        `Ready Inceptron model ${model.id} refers to missing base model ${model.models_dev.base_model}`,
+      );
+    }
+    validateReasoningContract(model as ReadyInceptronModel);
+    validatePricing(model);
+    validateModalities(model);
+    return model as ReadyInceptronModel;
+  });
+}
+
+export function buildInceptronModel(model: ReadyInceptronModel): SyncedModel {
+  const features = new Set(model.supported_features);
+  const samplingParameters = new Set(model.supported_sampling_parameters);
+  const input = validateModalities(model).input;
+  const output = validateModalities(model).output;
+  const limit = {
+    context: model.context_length,
+    output: model.max_output_length,
+  };
+
+  return factorBaseModel(
+    model.models_dev.base_model,
+    {
+      name: model.name,
+      attachment: input.some((modality) => modality !== "text"),
+      reasoning: features.has("reasoning"),
+      reasoning_options: model.models_dev.reasoning_options,
+      interleaved: model.models_dev.interleaved,
+      tool_call: features.has("tools"),
+      structured_output: features.has("structured_outputs"),
+      temperature: samplingParameters.has("temperature"),
+      status: model.models_dev.status,
+      cost: {
+        input: perTokenToPerMillion(model.pricing.prompt),
+        output: perTokenToPerMillion(model.pricing.completion),
+        cache_read: optionalPrice(model.pricing.input_cache_reads),
+        cache_write: optionalPrice(model.pricing.input_cache_writes),
+      },
+      limit,
+      modalities: { input, output },
+    },
+    limit,
+  );
+}
+
+function baseModelExists(modelID: string) {
+  return existsSync(path.join(MODELS_DIR, `${modelID}.toml`));
+}
+
+function validateReasoningContract(model: ReadyInceptronModel) {
+  const supportsReasoning = model.supported_features.includes("reasoning");
+  const options = model.models_dev.reasoning_options;
+  if (supportsReasoning !== (options !== undefined)) {
+    throw new Error(
+      `Inceptron model ${model.id} must expose reasoning_options exactly when reasoning is supported`,
+    );
+  }
+  if (model.models_dev.interleaved !== undefined && !supportsReasoning) {
+    throw new Error(`Inceptron model ${model.id} exposes interleaving without reasoning`);
+  }
+
+  const optionTypes = options?.map((option) => option.type) ?? [];
+  if (new Set(optionTypes).size !== optionTypes.length) {
+    throw new Error(`Inceptron model ${model.id} has duplicate reasoning option types`);
+  }
+  const exposesEffort = optionTypes.includes("effort");
+  const advertisesEffort = model.supported_sampling_parameters.includes("reasoning_effort");
+  if (exposesEffort !== advertisesEffort) {
+    throw new Error(
+      `Inceptron model ${model.id} must advertise reasoning_effort exactly when effort options are exposed`,
+    );
+  }
+}
+
+type Modality = "text" | "audio" | "image" | "video" | "pdf";
+const MODALITIES = new Set<Modality>(["text", "audio", "image", "video", "pdf"]);
+
+function validateModalities(model: InceptronModel): { input: Modality[]; output: Modality[] } {
+  const parse = (direction: "input" | "output", values: string[]) => {
+    const unique = [...new Set(values)];
+    for (const value of unique) {
+      if (!MODALITIES.has(value as Modality)) {
+        throw new Error(`Inceptron model ${model.id} has unsupported ${direction} modality: ${value}`);
+      }
+    }
+    return unique as Modality[];
+  };
+  return {
+    input: parse("input", model.input_modalities),
+    output: parse("output", model.output_modalities),
+  };
+}
+
+function validatePricing(model: InceptronModel) {
+  perTokenToPerMillion(model.pricing.prompt);
+  perTokenToPerMillion(model.pricing.completion);
+  optionalPrice(model.pricing.input_cache_reads);
+  optionalPrice(model.pricing.input_cache_writes);
+}
+
+function optionalPrice(value: string | undefined) {
+  return value === undefined ? undefined : perTokenToPerMillion(value);
+}
+
+/** Convert a non-negative decimal USD/token string to USD/million tokens without floating-point multiplication. */
+export function perTokenToPerMillion(value: string): number {
+  const match = /^(0|[1-9]\d*)(?:\.(\d+))?$/.exec(value);
+  if (match === null) throw new Error(`Invalid Inceptron per-token price: ${value}`);
+
+  const integer = match[1] as string;
+  const fraction = match[2] ?? "";
+  const digits = `${integer}${fraction}`.replace(/^0+(?=\d)/, "");
+  const decimalPlaces = fraction.length - 6;
+  const scaled = decimalPlaces <= 0
+    ? `${digits}${"0".repeat(-decimalPlaces)}`
+    : `${digits.slice(0, -decimalPlaces) || "0"}.${digits.slice(-decimalPlaces).padStart(decimalPlaces, "0")}`;
+  const result = Number(scaled);
+  if (!Number.isFinite(result) || result < 0) {
+    throw new Error(`Invalid Inceptron per-token price: ${value}`);
+  }
+  return result;
+}

packages/core/src/sync/providers/llmgateway.ts

@@ -2,6 +2,7 @@ import { z } from "zod";
 
 import { describeModel } from "../../describe.js";
 import { inferKimiFamily, ModelFamilyValues } from "../../family.js";
+import { ReasoningOption } from "../../schema.js";
 import type { ExistingModel, SyncProvider, SyncedFullModel, SyncedModel } from "../index.js";
 import { factorBaseModel, resolveCanonicalBaseModel } from "./openrouter.js";
 
@@ -11,12 +12,14 @@ const API_ENDPOINT = "https://api.llmgateway.io/v1/models";
 // canonical prefixes understood by resolveCanonicalBaseModel. Alias the few that
 // spell the lab differently. (Mirrors huggingface's CANONICAL_ORG_PREFIXES.)
 const CANONICAL_FAMILY_ALIASES: Record<string, string> = {
+  grok: "xai",
   mistral: "mistralai",
   moonshot: "moonshotai",
 };
 
 const BASE_MODEL_ALIASES: Record<string, string> = {
   "glm-5-2": "zhipuai/glm-5.2",
+  "grok-4-6": "xai/grok-4.6",
 };
 
 const Pricing = z.object({
@@ -27,6 +30,21 @@ const Pricing = z.object({
   input_cache_write: z.string().optional(),
 });
 
+const LLMGatewayProvider = z.object({
+  reasoning_efforts: z.array(z.string()).optional(),
+}).passthrough();
+
+const ReasoningEffortOrder = new Map([
+  "none",
+  "minimal",
+  "low",
+  "medium",
+  "high",
+  "xhigh",
+  "max",
+  "default",
+].map((effort, index) => [effort, index]));
+
 export const LLMGatewayModel = z.object({
   id: z.string(),
   name: z.string(),
@@ -37,6 +55,7 @@ export const LLMGatewayModel = z.object({
     output_modalities: z.array(z.string()),
   }),
   pricing: Pricing,
+  providers: z.array(LLMGatewayProvider),
   context_length: z.number(),
   supported_parameters: z.array(z.string()),
   structured_outputs: z.boolean().optional(),
@@ -138,12 +157,14 @@ export function buildLLMGatewayModel(
   const completion = price(model.pricing.completion);
   const reasoning = model.supported_parameters.includes("reasoning")
     || model.supported_parameters.includes("include_reasoning");
+  const reasoningOptions = llmGatewayReasoningOptions(model, existing);
   const context = model.context_length > 0
     ? model.context_length
     : existing?.limit?.context ?? model.context_length;
 
-  // The gateway is authoritative for the volatile, gateway-specific data — cost
-  // and served limits. Its supported_parameters / modalities are too noisy to
+  // The gateway is authoritative for the volatile, gateway-specific data — cost,
+  // served limits, and explicitly advertised reasoning efforts. Its
+  // supported_parameters / modalities are too noisy to
   // drive capability fields (it omits "tools" for flagship models yet lists
   // "temperature" for ones the catalog deliberately marks temperature=false),
   // so those stay curated: preserved from the existing entry (which, for a
@@ -183,6 +204,7 @@ export function buildLLMGatewayModel(
           modalities: existing.modalities,
         }),
         reasoning: existing.reasoning,
+        reasoning_options: reasoningOptions,
         temperature: existing.temperature,
         tool_call: existing.tool_call,
         structured_output: existing.structured_output,
@@ -218,6 +240,7 @@ export function buildLLMGatewayModel(
       last_updated: existing.last_updated ?? dateFromTimestamp(model.created),
       attachment: existing.attachment ?? false,
       reasoning: existing.reasoning ?? false,
+      reasoning_options: reasoningOptions,
       temperature: existing.temperature ?? false,
       tool_call: existing.tool_call ?? false,
       structured_output: existing.structured_output,
@@ -240,7 +263,11 @@ export function buildLLMGatewayModel(
   const canonical = resolveLLMGatewayBaseModel(model);
   if (canonical !== undefined) {
     const factoredLimit = { context, input: undefined, output: undefined };
-    return factorBaseModel(canonical, { limit: factoredLimit, cost }, factoredLimit);
+    return factorBaseModel(canonical, {
+      reasoning_options: reasoningOptions,
+      limit: factoredLimit,
+      cost,
+    }, factoredLimit);
   }
 
   // Brand-new model: best-effort translation from the gateway. Capability and
@@ -265,6 +292,7 @@ export function buildLLMGatewayModel(
     last_updated: dateFromTimestamp(model.created),
     attachment: input.some((value) => value !== "text"),
     reasoning,
+    reasoning_options: reasoningOptions,
     temperature: model.supported_parameters.includes("temperature"),
     tool_call: model.supported_parameters.includes("tools")
       || model.supported_parameters.includes("tool_choice"),
@@ -276,6 +304,27 @@ export function buildLLMGatewayModel(
   } satisfies SyncedFullModel;
 }
 
+function llmGatewayReasoningOptions(
+  model: LLMGatewayModel,
+  existing: ExistingModel | undefined,
+): SyncedFullModel["reasoning_options"] {
+  const advertised = new Set(model.providers.flatMap((provider) => provider.reasoning_efforts ?? []));
+  if (advertised.size === 0) return undefined;
+
+  const efforts = [...advertised].sort((a, b) => {
+    const order = (ReasoningEffortOrder.get(a) ?? Number.MAX_SAFE_INTEGER)
+      - (ReasoningEffortOrder.get(b) ?? Number.MAX_SAFE_INTEGER);
+    return order || a.localeCompare(b);
+  });
+  const preserved = existing?.reasoning_options?.filter((option) =>
+    option.type !== "effort" && !(option.type === "toggle" && advertised.has("none"))
+  ) ?? [];
+  return [
+    ...preserved,
+    ReasoningOption.parse({ type: "effort", values: efforts }),
+  ];
+}
+
 function defaultModalities(model: LLMGatewayModel) {
   return {
     input: modalities(model.architecture.input_modalities, ["text"]),

packages/core/src/sync/providers/merge-gateway.ts

@@ -165,7 +165,7 @@ export const mergeGateway = {
 export function mergeGatewayReasoningOptions(
   reasoning: MergeGatewayVendor["capabilities"]["reasoning"],
 ): NonNullable<SyncedFullModel["reasoning_options"]> | undefined {
-  if (reasoning === undefined) return undefined;
+  if (reasoning == null) return undefined;
   const options: NonNullable<SyncedFullModel["reasoning_options"]> = [];
 
   if (reasoning.disable_supported === true) {

packages/core/src/sync/providers/nano-gpt.ts

@@ -96,6 +96,7 @@ const BASE_MODEL_ALIASES: Record<string, string | undefined> = {
   "claude-opus-4": "anthropic/claude-opus-4-0",
   "claude-sonnet-4": "anthropic/claude-sonnet-4-0",
   "cohere/north-mini-code": "cohere/north-mini-code-1-0",
+  "doubao-seed-2-0-code-preview-260215": "bytedance-seed/seed-2.0-code",
 };
 
 const NANO_GPT_VARIANT_SUFFIX = /(?::(?:thinking|none|minimal|low|medium|high|xhigh|max|\d+)|-thinking)$/i;

packages/core/src/sync/providers/ofox.ts

@@ -55,8 +55,11 @@ export const ofox = {
   name: "Ofox",
   modelsDir: "providers/ofox/models",
   skipCreates: true,
-  trackMissingModels: false,
+  trackMissingModels: true,
   deleteMissing: false,
+  sourceID(model) {
+    return model.mode === "chat" ? model.id : undefined;
+  },
   missingNotice(paths) {
     return paths.map(
       (file) => `Ofox catalog no longer lists ${file}; review for manual deprecation or removal.`,

packages/core/src/sync/providers/openrouter.ts

@@ -13,6 +13,7 @@ const modelMetadataFilesByProvider = new Map<string, Set<string>>();
 let allModelMetadataIDs: string[] | undefined;
 
 const CANONICAL_BASE_MODEL_OVERRIDES = {
+  "bytedance/dola-seed-2.0-code": "bytedance-seed/seed-2.0-code",
   "openai/gpt-5.6-luna-pro": "openai/gpt-5.6-luna",
   "openai/gpt-5.6-sol-pro": "openai/gpt-5.6-sol",
   "openai/gpt-5.6-terra-pro": "openai/gpt-5.6-terra",
@@ -23,6 +24,7 @@ const CANONICAL_BASE_MODEL_OVERRIDES = {
 const CANONICAL_PROVIDER_PREFIXES = {
   alibaba: { provider: "alibaba", metadata: "alibaba" },
   anthropic: { provider: "anthropic", metadata: "anthropic" },
+  "bytedance-seed": { provider: "bytedance-seed", metadata: "bytedance-seed" },
   cohere: { provider: "cohere", metadata: "cohere" },
   deepseek: { provider: "deepseek", metadata: "deepseek" },
   google: { provider: "google", metadata: "google" },
@@ -319,6 +321,9 @@ function openRouterReasoningOptions(reasoning: OpenRouterModel["reasoning"]): Sy
     : reasoning.supported_efforts;
 
   if (efforts !== undefined) {
+    if (!reasoning.mandatory && !efforts.includes("none")) {
+      options.push({ type: "toggle" });
+    }
     options.push({
       type: "effort",
       values: reasoning.mandatory ? efforts.filter((value) => value !== "none") : [...efforts],

packages/core/src/sync/providers/pioneer.ts

@@ -141,9 +141,7 @@ export const pioneer = {
   name: "Pioneer",
   modelsDir: "providers/pioneer/models",
   skipCreates: true,
-  // Pioneer reports 2024-01-01 for every model, so its creation dates cannot
-  // support a meaningful age cutoff for remote-only model notifications.
-  trackMissingModels: false,
+  trackMissingModels: true,
   deleteMissing: false,
   async fetchModels() {
     const response = await fetch(API_ENDPOINT);

packages/core/test/auto-merge.test.ts

@@ -82,13 +82,28 @@ test("does not inspect deleted models", async () => {
 });
 
 test("allows reviewed providers with explicit reasoning options", async () => {
+  for (const provider of ["empiriolabs", "kilo", "llmgateway", "merge-gateway", "nano-gpt", "openrouter", "venice"]) {
+    const decision = await classifyAutoMerge(
+      [{ status: "updated", path: `providers/${provider}/models/reasoner.toml` }],
+      async () => fullModel(true, 'reasoning_options = [{ type = "toggle" }]'),
+      async () => fullModel(true, "reasoning_options = []"),
+    );
+
+    expect(decision.safe).toBe(true);
+  }
+});
+
+test("requires manual review for Inceptron reasoning updates", async () => {
   const decision = await classifyAutoMerge(
-    [{ status: "updated", path: "providers/openrouter/models/reasoner.toml" }],
-    async () => fullModel(true, 'reasoning_options = [{ type = "toggle" }]'),
+    [{ status: "updated", path: "providers/inceptron/models/reasoner.toml" }],
+    async () => fullModel(true, 'reasoning_options = [{ type = "effort", values = ["high"] }]'),
     async () => fullModel(true, "reasoning_options = []"),
   );
 
-  expect(decision.safe).toBe(true);
+  expect(decision.safe).toBe(false);
+  expect(decision.reasons).toContain(
+    "providers/inceptron/models/reasoner.toml is a reasoning model that requires manual review",
+  );
 });
 
 test("resolves reasoning from base model", async () => {

packages/core/test/empiriolabs.test.ts

@@ -0,0 +1,34 @@
+import { expect, test } from "bun:test";
+
+import { resolveEmpiriolabsBaseModel } from "../src/sync/providers/empiriolabs.js";
+
+test("resolves existing lab metadata without a hardcoded map", () => {
+  expect(resolveEmpiriolabsBaseModel("muse-glimmer-30b")).toBe("meta/muse-glimmer-30b");
+  expect(resolveEmpiriolabsBaseModel("muse-spark-1-2")).toBe("meta/muse-spark-1.2");
+  expect(resolveEmpiriolabsBaseModel("muse-spark-1-1")).toBe("meta/muse-spark-1.1");
+  expect(resolveEmpiriolabsBaseModel("seed-2-1-turbo")).toBe("bytedance-seed/seed-2.1-turbo");
+  expect(resolveEmpiriolabsBaseModel("seed-2-0-code")).toBe("bytedance-seed/seed-2.0-code");
+  expect(resolveEmpiriolabsBaseModel("seed-2-0-lite")).toBe("bytedance-seed/seed-2.0-lite");
+  expect(resolveEmpiriolabsBaseModel("seed-2-0-mini")).toBe("bytedance-seed/seed-2.0-mini");
+  expect(resolveEmpiriolabsBaseModel("seed-2-0-pro")).toBe("bytedance-seed/seed-2.0-pro");
+  expect(resolveEmpiriolabsBaseModel("qwen3-8-max")).toBe("alibaba/qwen3.8-max");
+});
+
+test("maps versioned slugs onto the undated canonical when needed", () => {
+  expect(resolveEmpiriolabsBaseModel("fugu-ultra-v1-1")).toBe("sakana/fugu-ultra");
+  expect(resolveEmpiriolabsBaseModel("fugu-ultra-v1-0")).toBe("sakana/fugu-ultra");
+});
+
+test("keeps true filename aliases", () => {
+  expect(resolveEmpiriolabsBaseModel("mistral-medium-3")).toBe("mistral/mistral-medium-2505");
+  expect(resolveEmpiriolabsBaseModel("mistral-small-4")).toBe("mistral/mistral-small-2603");
+});
+
+test("resolves a non-alias Mistral id through the mistralai prefix", () => {
+  expect(resolveEmpiriolabsBaseModel("mistral-small-2603")).toBe("mistral/mistral-small-2603");
+});
+
+test("does not invent lab metadata when none exists", () => {
+  expect(resolveEmpiriolabsBaseModel("deepreasoning")).toBeUndefined();
+  expect(resolveEmpiriolabsBaseModel("nova-pro-1-0")).toBeUndefined();
+});

packages/core/test/schema.test.ts

@@ -1,7 +1,7 @@
 import { describe, expect, test } from "bun:test";
 import { z } from "zod";
 
-import { AuthoredModel } from "../src/index.js";
+import { AuthoredModel, Provider } from "../src/index.js";
 
 type AuthoredModelData = z.infer<typeof AuthoredModel>;
 
@@ -109,6 +109,28 @@ describe("model schema", () => {
   });
 });
 
+describe("provider schema", () => {
+  const mergeGatewayProvider = {
+    id: "merge-gateway",
+    name: "Merge Gateway",
+    env: ["MERGE_GATEWAY_API_KEY"],
+    npm: "merge-gateway-ai-sdk-provider",
+    api: "https://api-gateway.merge.dev/v1/ai-sdk",
+    doc: "https://docs.merge.dev/merge-gateway",
+    models: {},
+  };
+
+  test("accepts Merge Gateway's native package with its OpenAI-compatible API", () => {
+    expect(Provider.safeParse(mergeGatewayProvider).success).toBe(true);
+  });
+
+  test("requires the compatibility API for the Merge Gateway package", () => {
+    const { api: _api, ...providerWithoutApi } = mergeGatewayProvider;
+
+    expect(Provider.safeParse(providerWithoutApi).success).toBe(false);
+  });
+});
+
 function baseModel(overrides: Partial<AuthoredModelData>) {
   return {
     id: "example/model",

packages/core/test/sync.test.ts

@@ -1,5 +1,5 @@
 import { expect, test } from "bun:test";
-import { mkdir, mkdtemp, readFile, rm } from "node:fs/promises";
+import { copyFile, mkdir, mkdtemp, readFile, rm } from "node:fs/promises";
 import { tmpdir } from "node:os";
 import path from "node:path";
 
@@ -13,9 +13,14 @@ import {
 import { buildCortecsModel, type CortecsModel } from "../src/sync/providers/cortecs.js";
 import {
   buildCrossModel,
+  CrossModelResponse,
   type CrossModelModel,
 } from "../src/sync/providers/crossmodel.js";
-import { buildDeepInfraModel, type DeepInfraModel } from "../src/sync/providers/deepinfra.js";
+import {
+  buildDeepInfraModel,
+  resolveDeepInfraBaseModel,
+  type DeepInfraModel,
+} from "../src/sync/providers/deepinfra.js";
 import {
   buildDigitalOceanModel,
   digitalocean,
@@ -24,7 +29,21 @@ import {
   resolveDigitalOceanBaseModel,
   type DigitalOceanSourceModel,
 } from "../src/sync/providers/digitalocean.js";
+import {
+  buildEdenAIModel,
+  collectFirstPartyBaseModels,
+  reasoningOptionsFor,
+  resolveEdenAIBaseModel,
+  type EdenAIModel,
+} from "../src/sync/providers/edenai.js";
 import { buildHyperModel, type HyperModel } from "../src/sync/providers/hyper.js";
+import {
+  buildInceptronModel,
+  parseInceptronModels,
+  perTokenToPerMillion,
+  type InceptronModel,
+  type ReadyInceptronModel,
+} from "../src/sync/providers/inceptron.js";
 import {
   buildEmpiriolabsModel,
   empiriolabs,
@@ -54,13 +73,14 @@ import {
   type NanoGptModel,
 } from "../src/sync/providers/nano-gpt.js";
 import { openai, parseOpenAIModels } from "../src/sync/providers/openai.js";
+import { ofox } from "../src/sync/providers/ofox.js";
 import { pioneer } from "../src/sync/providers/pioneer.js";
 import { google, shouldTrackGoogleModel } from "../src/sync/providers/google.js";
 import { buildTinfoilModel, tinfoil, type TinfoilModel } from "../src/sync/providers/tinfoil.js";
 import { resolveVeniceBaseModel } from "../src/sync/providers/venice.js";
 import { buildVercelModel, vercel } from "../src/sync/providers/vercel.js";
 import { buildWandbModel, type WandbModel } from "../src/sync/providers/wandb.js";
-import { buildXAIModel } from "../src/sync/providers/xai.js";
+import { buildXAIModel, xai } from "../src/sync/providers/xai.js";
 
 function anthropicModel(overrides: Partial<AnthropicModel> = {}): AnthropicModel {
   return {
@@ -145,6 +165,252 @@ function crossModelModel(overrides: Partial<CrossModelModel> = {}): CrossModelMo
   };
 }
 
+function inceptronModel(overrides: Partial<InceptronModel> = {}): InceptronModel {
+  return {
+    id: "zai-org/GLM-5.2",
+    name: "GLM 5.2",
+    context_length: 1_048_576,
+    max_output_length: 1_048_576,
+    input_modalities: ["text"],
+    output_modalities: ["text"],
+    supported_features: ["chat", "tools", "reasoning", "structured_outputs"],
+    supported_sampling_parameters: ["temperature", "reasoning_effort"],
+    pricing: {
+      prompt: "0.00000075",
+      completion: "0.0000029",
+      input_cache_reads: "0.00000017",
+      input_cache_writes: "0",
+    },
+    models_dev: {
+      base_model: "zhipuai/glm-5.2",
+      reasoning_options: [{ type: "effort", values: ["high", "max"] }],
+      interleaved: { field: "reasoning_content" },
+      status: "alpha",
+    },
+    ...overrides,
+  };
+}
+
+function readyInceptronModel(overrides: Partial<InceptronModel> = {}): ReadyInceptronModel {
+  return parseInceptronModels({
+    object: "list",
+    data: [inceptronModel(overrides)],
+  })[0]!;
+}
+
+test("builds current Inceptron models from explicit base metadata", () => {
+  const models = parseInceptronModels({
+    object: "list",
+    data: [
+      inceptronModel({
+        id: "MiniMaxAI/MiniMax-M2.5",
+        name: "MiniMax M2.5",
+        context_length: 196_608,
+        max_output_length: 196_608,
+        pricing: { prompt: "0.00000022", completion: "0.0000009" },
+        models_dev: {
+          base_model: "minimax/MiniMax-M2.5",
+          reasoning_options: [{ type: "effort", values: ["low", "medium", "high"] }],
+        },
+      }),
+      inceptronModel(),
+      inceptronModel({
+        id: "moonshotai/Kimi-K2.6",
+        name: "Kimi K2.6",
+        context_length: 262_144,
+        max_output_length: 262_144,
+        input_modalities: ["text", "image"],
+        supported_sampling_parameters: ["temperature"],
+        pricing: { prompt: "0.0000006", completion: "0.00000341" },
+        models_dev: {
+          base_model: "moonshotai/kimi-k2.6",
+          reasoning_options: [],
+          interleaved: { field: "reasoning_content" },
+        },
+      }),
+      inceptronModel({
+        id: "moonshotai/Kimi-K2.7-Code",
+        name: "Kimi K2.7 Code",
+        context_length: 262_144,
+        max_output_length: 262_144,
+        input_modalities: ["text", "image"],
+        supported_sampling_parameters: ["temperature"],
+        pricing: { prompt: "0.0000007", completion: "0.0000035" },
+        models_dev: {
+          base_model: "moonshotai/kimi-k2.7-code",
+          reasoning_options: [],
+          interleaved: { field: "reasoning_content" },
+        },
+      }),
+      inceptronModel({
+        id: "deepseek-ai/DeepSeek-V4-Flash-0731",
+        name: "DeepSeek V4 Flash 0731",
+        context_length: 1_048_576,
+        max_output_length: 1_048_576,
+        pricing: {
+          prompt: "0.00000013",
+          completion: "0.00000028",
+          input_cache_reads: "0.00000003",
+          input_cache_writes: "0",
+        },
+        models_dev: {
+          base_model: "deepseek/deepseek-v4-flash-0731",
+          reasoning_options: [{ type: "effort", values: ["high", "max"] }],
+          interleaved: { field: "reasoning_content" },
+        },
+      }),
+    ],
+  });
+
+  const built = models.map(buildInceptronModel);
+  expect(built.map((model) => "base_model" in model ? model.base_model : undefined)).toEqual([
+    "minimax/MiniMax-M2.5",
+    "zhipuai/glm-5.2",
+    "moonshotai/kimi-k2.6",
+    "moonshotai/kimi-k2.7-code",
+    "deepseek/deepseek-v4-flash-0731",
+  ]);
+  expect(built[0]).toMatchObject({
+    reasoning_options: [{ type: "effort", values: ["low", "medium", "high"] }],
+    cost: { input: 0.22, output: 0.9 },
+  });
+  expect(built[1]).toMatchObject({
+    name: "GLM 5.2",
+    reasoning_options: [{ type: "effort", values: ["high", "max"] }],
+    interleaved: { field: "reasoning_content" },
+    status: "alpha",
+    cost: { input: 0.75, output: 2.9, cache_read: 0.17, cache_write: 0 },
+    limit: { context: 1_048_576, output: 1_048_576 },
+  });
+  expect(built[2]).toMatchObject({
+    reasoning_options: [],
+    interleaved: { field: "reasoning_content" },
+    modalities: { input: ["text", "image"] },
+  });
+  expect(built[3]).toMatchObject({
+    reasoning_options: [],
+    interleaved: { field: "reasoning_content" },
+  });
+  expect(built[4]).toMatchObject({
+    reasoning_options: [{ type: "effort", values: ["high", "max"] }],
+    interleaved: { field: "reasoning_content" },
+    cost: { input: 0.13, output: 0.28, cache_read: 0.03, cache_write: 0 },
+    limit: { context: 1_048_576, output: 1_048_576 },
+  });
+});
+
+test("converts Inceptron per-token decimal prices exactly", () => {
+  expect(perTokenToPerMillion("0")).toBe(0);
+  expect(perTokenToPerMillion("0.00000005")).toBe(0.05);
+  expect(perTokenToPerMillion("0.00000341")).toBe(3.41);
+  expect(perTokenToPerMillion("1.25")).toBe(1_250_000);
+});
+
+test("rejects incomplete or contradictory ready Inceptron catalogs", () => {
+  expect(() =>
+    parseInceptronModels({ object: "list", data: [inceptronModel({ models_dev: undefined })] })
+  ).toThrow("missing models_dev metadata");
+  expect(() =>
+    parseInceptronModels({
+      object: "list",
+      data: [inceptronModel(), inceptronModel()],
+    })
+  ).toThrow("Duplicate ready Inceptron model ID");
+  expect(() =>
+    readyInceptronModel({
+      models_dev: { base_model: "zhipuai/not-a-real-model", reasoning_options: [] },
+      supported_sampling_parameters: [],
+    })
+  ).toThrow("missing base model");
+  expect(() =>
+    readyInceptronModel({ pricing: { prompt: "1e-6", completion: "0.1" } })
+  ).toThrow("Invalid Inceptron per-token price");
+  expect(() => readyInceptronModel({ input_modalities: ["text", "binary"] }))
+    .toThrow("unsupported input modality");
+  expect(() =>
+    readyInceptronModel({
+      models_dev: { base_model: "zhipuai/glm-5.2", reasoning_options: [] },
+    })
+  ).toThrow("reasoning_effort exactly when effort options are exposed");
+});
+
+test("ignores not-ready Inceptron models while validating every ready model", () => {
+  const ready = inceptronModel();
+  const notReady = inceptronModel({
+    id: "staged/model",
+    is_ready: false,
+    models_dev: undefined,
+    input_modalities: ["unsupported-but-ignored"],
+    pricing: { prompt: "malformed", completion: "malformed" },
+  });
+  expect(parseInceptronModels({ object: "list", data: [ready, notReady] })).toHaveLength(1);
+
+  expect(() =>
+    parseInceptronModels({
+      object: "list",
+      data: [ready, inceptronModel({ id: "ready/model", models_dev: undefined })],
+    })
+  ).toThrow("missing models_dev metadata");
+});
+
+test("syncs authoritative Inceptron additions, updates, and removals", async () => {
+  const root = await mkdtemp(path.join(tmpdir(), "models-dev-inceptron-"));
+  const modelsDir = path.join(root, "providers", "inceptron", "models");
+  await mkdir(modelsDir, { recursive: true });
+  for (const base of ["zhipuai/glm-5.2", "moonshotai/kimi-k2.6"]) {
+    const destination = path.join(root, "models", `${base}.toml`);
+    await mkdir(path.dirname(destination), { recursive: true });
+    await copyFile(path.join(import.meta.dirname, "..", "..", "..", "models", `${base}.toml`), destination);
+  }
+
+  let source: ReadyInceptronModel[] = [
+    readyInceptronModel(),
+    readyInceptronModel({
+      id: "moonshotai/Kimi-K2.6",
+      name: "Kimi K2.6",
+      context_length: 262_144,
+      max_output_length: 262_144,
+      input_modalities: ["text", "image"],
+      supported_sampling_parameters: ["temperature"],
+      models_dev: {
+        base_model: "moonshotai/kimi-k2.6",
+        reasoning_options: [],
+        interleaved: true,
+      },
+    }),
+  ];
+  const provider: SyncProvider<ReadyInceptronModel> = {
+    id: "inceptron-test",
+    name: "Inceptron test",
+    modelsDir,
+    async fetchModels() {
+      return source;
+    },
+    parseModels(raw) {
+      return raw as ReadyInceptronModel[];
+    },
+    translateModel(model) {
+      return { id: model.id, model: buildInceptronModel(model) };
+    },
+  };
+
+  try {
+    const initial = await syncProvider(provider);
+    expect(initial).toMatchObject({ created: 2, updated: 0, deleted: 0 });
+
+    source = [readyInceptronModel({
+      pricing: { prompt: "0.0000008", completion: "0.0000029" },
+    })];
+    const changed = await syncProvider(provider);
+    expect(changed).toMatchObject({ created: 0, updated: 1, deleted: 1 });
+
+    const unchanged = await syncProvider(provider);
+    expect(unchanged).toMatchObject({ created: 0, updated: 0, deleted: 0, unchanged: 1 });
+  } finally {
+    await rm(root, { recursive: true, force: true });
+  }
+});
+
 test("syncs CrossModel's structured-output capability", () => {
   const supported = buildCrossModel(crossModelModel(), undefined);
   const unsupported = buildCrossModel(
@@ -176,6 +442,31 @@ test("syncs CrossModel's structured-output capability", () => {
   });
 });
 
+test("parses CrossModel's nullable reasoning controls", () => {
+  const parsed = CrossModelResponse.parse({
+    data: [
+      {
+        ...crossModelModel(),
+        capabilities: {
+          reasoning: {
+            supported: true,
+            toggle: null,
+            effort: null,
+            budget_tokens: null,
+          },
+        },
+      },
+    ],
+  });
+
+  expect(parsed.data[0]?.capabilities?.reasoning).toEqual({
+    supported: true,
+    toggle: undefined,
+    effort: undefined,
+    budget_tokens: undefined,
+  });
+});
+
 test("syncs NanoGPT's verified reasoning, pricing, limits, and open-weight metadata", () => {
   const model = buildNanoGptModel(nanoGptModel({
     pricing: {
@@ -273,6 +564,8 @@ test("factors NanoGPT variants against canonical models without retaining wrong
   expect(resolveNanoGptBaseModel("TEE/gpt-oss-120b")).toBe("openai/gpt-oss-120b");
   expect(resolveNanoGptBaseModel("TEE/gemma-4-31b-it")).toBe("google/gemma-4-31b-it");
   expect(resolveNanoGptBaseModel("cohere/north-mini-code")).toBe("cohere/north-mini-code-1-0");
+  expect(resolveNanoGptBaseModel("doubao-seed-2-0-code-preview-260215"))
+    .toBe("bytedance-seed/seed-2.0-code");
   expect(resolveNanoGptBaseModel("xiaomi/mimo-v2.5-pro-ultraspeed"))
     .toBe("xiaomi/mimo-v2.5-pro-ultraspeed");
   expect(resolveNanoGptBaseModel("claude-haiku-4-5-20251001-thinking"))
@@ -802,13 +1095,19 @@ test("OpenAI availability sync retains models absent from a scoped response", as
   }
 });
 
-test("does not track unreliable remote-only models", () => {
+test("tracks missing models except for unreliable first-party inventories", () => {
   expect(google.skipCreates).toBe(true);
   expect(google.trackMissingModels).toBe(false);
   expect(openai.skipCreates).toBe(true);
   expect(openai.trackMissingModels).toBe(false);
   expect(pioneer.skipCreates).toBe(true);
-  expect(pioneer.trackMissingModels).toBe(false);
+  expect(pioneer.trackMissingModels).toBe(true);
+  expect(ofox.skipCreates).toBe(true);
+  expect(ofox.trackMissingModels).toBe(true);
+  expect(tinfoil.skipCreates).toBe(true);
+  expect(tinfoil.trackMissingModels).not.toBe(false);
+  expect(xai.skipCreates).toBe(true);
+  expect(xai.trackMissingModels).not.toBe(false);
 });
 
 test("tracks public Google model families but not opaque internal IDs", () => {
@@ -1946,6 +2245,171 @@ test("factors new Hyper models against unique models/ metadata", () => {
   });
 });
 
+test("factors Eden AI models onto lab metadata and prices from list_pricing", () => {
+  const model = edenAIModel({
+    id: "openai/gpt-5.6-terra",
+    model_name: "gpt-5.6-terra",
+    owned_by: "openai",
+    pricing: { input_cost_per_token: 0.0000013, output_cost_per_token: 0.0000078 },
+    list_pricing: {
+      input_cost_per_token: 0.000002,
+      output_cost_per_token: 0.000012,
+      cache_read_input_token_cost: 0.0000002,
+    },
+  });
+
+  expect(buildEdenAIModel(model)).toMatchObject({
+    base_model: "openai/gpt-5.6-terra",
+    cost: { input: 2, output: 12, cache_read: 0.2 },
+    reasoning_options: [
+      { type: "effort", values: ["none", "low", "medium", "high", "xhigh", "max"] },
+    ],
+  });
+  expect(buildEdenAIModel(model)).not.toHaveProperty("reasoning");
+});
+
+test("takes Eden AI reasoning options from the model's own lab entry", () => {
+  expect(reasoningOptionsFor("deepseek/deepseek-v4-pro")).toEqual([
+    { type: "effort", values: ["none", "high", "max"] },
+  ]);
+  expect(reasoningOptionsFor("openai/o1")).toEqual([
+    { type: "effort", values: ["low", "medium", "high"] },
+  ]);
+});
+
+test("skips Eden AI models whose reasoning control has no effort equivalent", () => {
+  // Lab and OpenRouter both expose these through budget_tokens, which Eden AI
+  // has no request field for.
+  expect(reasoningOptionsFor("google/gemini-2.5-pro")).toBeUndefined();
+  expect(
+    buildEdenAIModel(
+      edenAIModel({
+        id: "google/gemini-2.5-pro",
+        model_name: "gemini-2.5-pro",
+        owned_by: "google",
+      }),
+    ),
+  ).toBeUndefined();
+});
+
+test("omits Eden AI reasoning options for non-reasoning models", () => {
+  const model = edenAIModel({
+    id: "openai/gpt-4o-mini",
+    model_name: "gpt-4o-mini",
+    owned_by: "openai",
+    context_length: 128_000,
+  });
+
+  const built = buildEdenAIModel(model);
+  expect(built).toMatchObject({ base_model: "openai/gpt-4o-mini" });
+  expect(built).not.toHaveProperty("reasoning_options");
+});
+
+test("skips Eden AI models without lab metadata", () => {
+  expect(
+    buildEdenAIModel(
+      edenAIModel({
+        id: "deepinfra/acme/Not-A-Real-Model",
+        model_name: "acme/Not-A-Real-Model",
+        owned_by: "deepinfra",
+      }),
+    ),
+  ).toBeUndefined();
+});
+
+test("names Eden AI regional deployments after the canonical model", () => {
+  expect(
+    buildEdenAIModel(
+      edenAIModel({
+        id: "amazon/anthropic.claude-opus-5@eu",
+        model_name: "anthropic.claude-opus-5",
+        owned_by: "amazon",
+      }),
+    ),
+  ).toMatchObject({
+    base_model: "anthropic/claude-opus-5",
+    name: "Claude Opus 5 (EU)",
+  });
+});
+
+test("builds Eden AI context tiers without reading time-based cache keys", () => {
+  const model = edenAIModel({
+    id: "openai/gpt-5.6-terra",
+    model_name: "gpt-5.6-terra",
+    owned_by: "openai",
+    list_pricing: {
+      input_cost_per_token: 0.000002,
+      output_cost_per_token: 0.000012,
+      input_cost_per_token_above_272k_tokens: 0.000004,
+      output_cost_per_token_above_272k_tokens: 0.000018,
+      cache_creation_input_token_cost_above_1hr: 0.000009,
+      cache_creation_input_token_cost_above_1hr_above_272k_tokens: 0.00001,
+    },
+  });
+
+  expect(buildEdenAIModel(model)).toMatchObject({
+    cost: {
+      input: 2,
+      output: 12,
+      tiers: [{ tier: { type: "context", size: 272_000 }, input: 4, output: 18 }],
+    },
+  });
+  expect(
+    (buildEdenAIModel(model) as { cost: { tiers: Array<Record<string, unknown>> } }).cost.tiers[0],
+  ).not.toHaveProperty("cache_write");
+});
+
+test("keeps only the first-party Eden AI route when the lab's own API is relayed", () => {
+  const bedrock = edenAIModel({
+    id: "amazon/anthropic.claude-opus-5",
+    model_name: "anthropic.claude-opus-5",
+    owned_by: "amazon",
+  });
+  const direct = edenAIModel({
+    id: "anthropic/claude-opus-5",
+    model_name: "claude-opus-5",
+    owned_by: "anthropic",
+  });
+
+  const firstParty = collectFirstPartyBaseModels([bedrock, direct]);
+  expect(firstParty).toEqual(new Set(["anthropic/claude-opus-5"]));
+  expect(buildEdenAIModel(bedrock, firstParty)).toBeUndefined();
+  expect(buildEdenAIModel(direct, firstParty)).toMatchObject({
+    base_model: "anthropic/claude-opus-5",
+  });
+});
+
+test("keeps every Eden AI route for models with no first-party relay", () => {
+  const models = ["deepinfra", "groq", "cerebras"].map((owner) =>
+    edenAIModel({
+      id: `${owner}/openai/gpt-oss-120b`,
+      model_name: "openai/gpt-oss-120b",
+      owned_by: owner,
+    }),
+  );
+
+  const firstParty = collectFirstPartyBaseModels(models);
+  expect(firstParty.size).toBe(0);
+  for (const model of models) {
+    expect(buildEdenAIModel(model, firstParty)).toMatchObject({
+      base_model: "openai/gpt-oss-120b",
+    });
+  }
+});
+
+test("resolves Eden AI aliases to the model they point at", () => {
+  expect(
+    resolveEdenAIBaseModel(
+      edenAIModel({
+        id: "anthropic/claude-opus-latest",
+        model_name: "claude-opus-latest",
+        owned_by: "anthropic",
+        alias_of: "anthropic/claude-opus-5",
+      }),
+    ),
+  ).toBe("anthropic/claude-opus-5");
+});
+
 test("formats interleaved as a root field before reasoning option tables", () => {
   const content = formatToml({
     id: "example/model",
@@ -2037,6 +2501,11 @@ test("formats provider overrides and experimental modes", () => {
   });
 });
 
+test("resolves DeepInfra ByteDance IDs to canonical metadata", () => {
+  expect(resolveDeepInfraBaseModel("ByteDance/Seed-2.0-code"))
+    .toBe("bytedance-seed/seed-2.0-code");
+});
+
 test("DeepInfra preserves live modalities for new base models", () => {
   const model = buildDeepInfraModel(
     deepInfraModel("Qwen/Qwen3.5-9B", ["multimodal", "input-video"]),
@@ -2160,6 +2629,29 @@ test("preserves authored Cortecs reasoning options missing from the API", () =>
   });
 });
 
+test("overrides canonical metadata with Cortecs reasoning support", () => {
+  const model: CortecsModel = {
+    id: "apertus-70b",
+    created: 1_775_088_000,
+    pricing: { currency: "EUR", input_token: 1.25, output_token: 2 },
+    context_size: 65_536,
+    input_modalities: ["text"],
+    output_modalities: ["text"],
+    supported_features: ["reasoning", "tools"],
+  };
+  const existing: ExistingModel = {
+    base_model: "swiss-ai/apertus-70b",
+    reasoning: true,
+    reasoning_options: [],
+  };
+
+  expect(buildCortecsModel(model, existing, existing)).toMatchObject({
+    base_model: "swiss-ai/apertus-70b",
+    reasoning: true,
+    reasoning_options: [],
+  });
+});
+
 test("syncs OpenRouter reasoning efforts from model metadata", () => {
   const model = buildOpenRouterModel(openRouterModel({
     reasoning: {
@@ -2171,6 +2663,7 @@ test("syncs OpenRouter reasoning efforts from model metadata", () => {
   expect(model).toMatchObject({
     base_model: "anthropic/claude-sonnet-5",
     reasoning_options: [
+      { type: "toggle" },
       { type: "effort", values: ["max", "xhigh", "high", "medium", "low"] },
     ],
   });
@@ -2229,17 +2722,23 @@ test("factors OpenRouter Pro routes against canonical OpenAI metadata", () => {
 // Ensures Merge Gateway namespaces reuse the matching canonical model metadata.
 test("resolves Merge Gateway provider aliases to canonical metadata", () => {
   expect([
+    resolveCanonicalBaseModel("bytedance-seed/seed-2.0-code"),
+    resolveCanonicalBaseModel("bytedance/dola-seed-2.0-code"),
     resolveCanonicalBaseModel("moonshot/kimi-k2.5"),
     resolveCanonicalBaseModel("moonshot/kimi-k2.6"),
     resolveCanonicalBaseModel("moonshot/kimi-k2.7-code"),
     resolveCanonicalBaseModel("moonshot/kimi-k2.7-code-highspeed"),
     resolveCanonicalBaseModel("sakana/fugu-ultra"),
+    resolveCanonicalBaseModel("meta/muse-glimmer-30b"),
   ]).toEqual([
+    "bytedance-seed/seed-2.0-code",
+    "bytedance-seed/seed-2.0-code",
     "moonshotai/kimi-k2.5",
     "moonshotai/kimi-k2.6",
     "moonshotai/kimi-k2.7-code",
     "moonshotai/kimi-k2.7-code-highspeed",
     "sakana/fugu-ultra",
+    "meta/muse-glimmer-30b",
   ]);
 });
 
@@ -2282,6 +2781,7 @@ test("prefers OpenRouter API reasoning options over authored ones", () => {
 
   expect(model).toMatchObject({
     reasoning_options: [
+      { type: "toggle" },
       { type: "effort", values: ["max", "xhigh", "high", "medium", "low"] },
     ],
   });
@@ -2334,6 +2834,7 @@ test("upgrades empty OpenRouter reasoning options from model metadata", () => {
 
   expect(model).toMatchObject({
     reasoning_options: [
+      { type: "toggle" },
       { type: "effort", values: ["high", "medium", "low"] },
     ],
   });
@@ -2355,6 +2856,62 @@ test("factors new LLM Gateway models against the canonical base metadata", () =>
   expect("modalities" in model).toBe(false);
 });
 
+test("syncs explicitly advertised LLM Gateway reasoning efforts", () => {
+  const model = buildLLMGatewayModel(llmGatewayModel({
+    id: "seed-2-1-turbo",
+    name: "Seed 2.1 Turbo",
+    family: "bytedance",
+    providers: [{
+      reasoning_efforts: ["high", "none", "max", "low", "xhigh", "minimal", "medium"],
+    }],
+  }), undefined);
+
+  expect(model).toMatchObject({
+    reasoning_options: [{
+      type: "effort",
+      values: ["none", "minimal", "low", "medium", "high", "xhigh", "max"],
+    }],
+  });
+});
+
+test("unions LLM Gateway reasoning efforts in canonical order", () => {
+  const model = buildLLMGatewayModel(llmGatewayModel({
+    id: "unreviewed-reasoner",
+    providers: [
+      { reasoning_efforts: ["high", "low"] },
+      { reasoning_efforts: ["none", "low", "xhigh"] },
+    ],
+  }), undefined);
+
+  expect(model).toMatchObject({
+    reasoning_options: [{
+      type: "effort",
+      values: ["none", "low", "high", "xhigh"],
+    }],
+  });
+});
+
+test("keeps non-effort LLM Gateway controls when syncing efforts", () => {
+  const model = buildLLMGatewayModel(llmGatewayModel({
+    providers: [{ reasoning_efforts: ["none", "low", "high"] }],
+  }), {
+    name: "Claude Fable 5",
+    reasoning: true,
+    reasoning_options: [
+      { type: "toggle" },
+      { type: "budget_tokens", min: 1024 },
+      { type: "effort", values: ["low"] },
+    ],
+  });
+
+  expect(model).toMatchObject({
+    reasoning_options: [
+      { type: "budget_tokens", min: 1024 },
+      { type: "effort", values: ["none", "low", "high"] },
+    ],
+  });
+});
+
 test("factors aliased LLM Gateway routes against canonical metadata", () => {
   const model = buildLLMGatewayModel(llmGatewayModel({
     id: "glm-5-2",
@@ -2381,6 +2938,29 @@ test("factors aliased LLM Gateway routes against canonical metadata", () => {
   });
 });
 
+test("factors Grok LLM Gateway routes against xAI metadata", () => {
+  const model = buildLLMGatewayModel(llmGatewayModel({
+    id: "grok-4-6",
+    name: "Grok 4.6",
+    family: "grok",
+    context_length: 500_000,
+    pricing: {
+      prompt: "2e-6",
+      completion: "6e-6",
+      input_cache_read: "0.5e-6",
+    },
+  }), undefined);
+
+  expect(model).toEqual({
+    base_model: "xai/grok-4.6",
+    cost: {
+      input: 2,
+      output: 6,
+      cache_read: 0.5,
+    },
+  });
+});
+
 // Ensures catalog pagination preserves authentication and returns every page.
 test("fetches every page of the Merge Gateway catalog", async () => {
   const requests: string[] = [];
@@ -2559,6 +3139,23 @@ test("confirms reasoning when any available Merge Gateway route reports supports
   expect(model).not.toMatchObject({ reasoning: false });
 });
 
+// The live catalog emits reasoning: null on some routes even when
+// supports_reasoning is true. Treat that as unknown controls, not a crash.
+test("tolerates a null Merge Gateway reasoning object when reasoning is confirmed", () => {
+  const selected = mergeGatewayVendor();
+  selected.capabilities.supports_reasoning = true;
+  selected.capabilities.reasoning = null;
+  const model = buildMergeGatewayModel(mergeGatewayModel({
+    vendors: { openai: selected },
+  }), {
+    base_model: "openai/gpt-5.6-sol",
+    cost: { input: 5, output: 30 },
+  });
+
+  expect(model).toMatchObject({ reasoning_options: [] });
+  expect(model).not.toMatchObject({ reasoning: false });
+});
+
 // Publishes a toggle only when the selected route explicitly supports disabling reasoning.
 test("derives a Merge Gateway reasoning toggle when the selected route supports disabling", () => {
   const selected = mergeGatewayVendor();
@@ -3139,6 +3736,7 @@ test("syncs EmpirioLabs pricing tiers and reasoning controls", () => {
 
 test("maps EmpirioLabs aliases to canonical model metadata", () => {
   expect(resolveEmpiriolabsBaseModel("fugu-ultra")).toBe("sakana/fugu-ultra");
+  expect(resolveEmpiriolabsBaseModel("seed-2-0-code")).toBe("bytedance-seed/seed-2.0-code");
   expect(resolveEmpiriolabsBaseModel("muse-spark-1-1")).toBe("meta/muse-spark-1.1");
   expect(resolveEmpiriolabsBaseModel("step-3-5-flash")).toBe("stepfun/step-3.5-flash");
 });
@@ -3171,6 +3769,7 @@ function llmGatewayModel(overrides: Partial<LLMGatewayModel> = {}): LLMGatewayMo
       input_cache_write: "12.5e-6",
       internal_reasoning: "0",
     },
+    providers: [{}],
     context_length: 1_000_000,
     supported_parameters: ["temperature", "max_tokens", "top_p", "effort", "reasoning"],
     structured_outputs: true,
@@ -3216,6 +3815,26 @@ function mergeGatewayModel(overrides: Partial<MergeGatewayModel> = {}): MergeGat
   };
 }
 
+function edenAIModel(overrides: Partial<EdenAIModel> = {}): EdenAIModel {
+  return {
+    id: "openai/gpt-5.6-terra",
+    owned_by: "openai",
+    model_name: "gpt-5.6-terra",
+    context_length: 1_050_000,
+    capabilities: {
+      input_modalities: ["text", "image"],
+      output_modalities: ["text"],
+      supports_function_calling: true,
+      supports_response_schema: true,
+    },
+    list_pricing: {
+      input_cost_per_token: 0.000002,
+      output_cost_per_token: 0.000012,
+    },
+    ...overrides,
+  };
+}
+
 function hyperModel(overrides: Partial<HyperModel> = {}): HyperModel {
   return {
     id: "deepseek-v4-flash",

packages/sdk/src/generated.ts

@@ -18,6 +18,7 @@ export type ModelFamily =
   | "claude"
   | "claude-fable"
   | "claude-haiku"
+  | "claude-mythos"
   | "claude-opus"
   | "claude-sonnet"
   | "codestral"
@@ -172,6 +173,7 @@ export type ModelFamily =
   | "qwen3.6"
   | "qwen3.7-max"
   | "qwen3.7-plus"
+  | "qwen3.8-max"
   | "qwerky"
   | "ray"
   | "recraft"

providers/abacus/models/claude-opus-5.toml

@@ -0,0 +1,18 @@
+# Source: https://routellm.abacus.ai/v1/models
+# RouteLLM input_modalities: ['text', 'image']
+# RouteLLM output_modalities: ['text']
+# RouteLLM context_length: 1000000
+# RouteLLM max_completion_tokens: 128000
+# RouteLLM input_token_rate: 0.000005
+# RouteLLM output_token_rate: 0.000025
+# RouteLLM thinking: True
+base_model = "anthropic/claude-opus-5"
+reasoning_options = []
+
+[cost]
+input = 5
+output = 25
+
+[modalities]
+input = ["text", "image"]
+output = ["text"]

providers/abacus/models/gemini-3-pro-image.toml

@@ -0,0 +1,15 @@
+# Source: https://routellm.abacus.ai/v1/models
+# RouteLLM input_modalities: ['text', 'image']
+# RouteLLM output_modalities: ['text', 'image']
+# RouteLLM context_length: 65536
+# RouteLLM max_completion_tokens: 32768
+# RouteLLM input_token_rate: 0.000002
+# RouteLLM output_token_rate: 0.000012
+# RouteLLM cached_input_token_rate: 0.0000002
+base_model = "google/gemini-3-pro-image"
+reasoning_options = []
+
+[cost]
+input = 2
+output = 12
+cache_read = 0.2

providers/abacus/models/gemini-3.1-flash-image.toml

@@ -0,0 +1,20 @@
+# Source: https://routellm.abacus.ai/v1/models
+# RouteLLM input_modalities: ['text', 'image']
+# RouteLLM output_modalities: ['text', 'image']
+# RouteLLM context_length: 1048576
+# RouteLLM max_completion_tokens: 32768
+# RouteLLM input_token_rate: 0.0000005
+# RouteLLM output_token_rate: 0.000003
+base_model = "google/gemini-3.1-flash-image"
+reasoning_options = []
+
+[cost]
+input = 0.5
+output = 3
+
+[limit]
+context = 1_048_576
+
+[modalities]
+input = ["text", "image"]
+output = ["text", "image"]

providers/abacus/models/gemini-3.5-flash-lite.toml

@@ -0,0 +1,20 @@
+# Source: https://routellm.abacus.ai/v1/models
+# RouteLLM input_modalities: ['text', 'image', 'audio']
+# RouteLLM output_modalities: ['text']
+# RouteLLM context_length: 1048576
+# RouteLLM max_completion_tokens: 65536
+# RouteLLM input_token_rate: 0.0000003
+# RouteLLM output_token_rate: 0.0000025
+# RouteLLM cached_input_token_rate: 0.00000003
+# RouteLLM thinking: True
+base_model = "google/gemini-3.5-flash-lite"
+reasoning_options = []
+
+[cost]
+input = 0.3
+output = 2.5
+cache_read = 0.03
+
+[modalities]
+input = ["text", "image", "audio"]
+output = ["text"]

providers/abacus/models/gemini-3.6-flash.toml

@@ -0,0 +1,20 @@
+# Source: https://routellm.abacus.ai/v1/models
+# RouteLLM input_modalities: ['text', 'image', 'audio']
+# RouteLLM output_modalities: ['text']
+# RouteLLM context_length: 1048576
+# RouteLLM max_completion_tokens: 65536
+# RouteLLM input_token_rate: 0.0000015
+# RouteLLM output_token_rate: 0.0000075
+# RouteLLM cached_input_token_rate: 0.00000015
+# RouteLLM thinking: True
+base_model = "google/gemini-3.6-flash"
+reasoning_options = []
+
+[cost]
+input = 1.5
+output = 7.5
+cache_read = 0.15
+
+[modalities]
+input = ["text", "image", "audio"]
+output = ["text"]

providers/abacus/models/gemini-3.7-flash.toml

@@ -0,0 +1,20 @@
+# Source: https://routellm.abacus.ai/v1/models
+# RouteLLM input_modalities: ['text', 'image', 'audio', 'video']
+# RouteLLM output_modalities: ['text']
+# RouteLLM context_length: 1048576
+# RouteLLM max_completion_tokens: 65536
+# RouteLLM input_token_rate: 0.00000075
+# RouteLLM output_token_rate: 0.00000375
+# RouteLLM cached_input_token_rate: 0.000000075
+# RouteLLM thinking: True
+base_model = "google/gemini-3.7-flash"
+reasoning_options = []
+
+[cost]
+input = 0.75
+output = 3.75
+cache_read = 0.075
+
+[modalities]
+input = ["text", "image", "audio", "video"]
+output = ["text"]

providers/abacus/models/grok-4.6.toml

@@ -0,0 +1,19 @@
+# Source: https://routellm.abacus.ai/v1/models
+# RouteLLM input_modalities: ['text', 'image']
+# RouteLLM output_modalities: ['text']
+# RouteLLM context_length: 500000
+# RouteLLM max_completion_tokens: 32768
+# RouteLLM input_token_rate: 0.000002
+# RouteLLM output_token_rate: 0.000006
+# RouteLLM cached_input_token_rate: 0.0000005
+# RouteLLM thinking: True
+base_model = "xai/grok-4.6"
+reasoning_options = []
+
+[cost]
+input = 2
+output = 6
+cache_read = 0.5
+
+[limit]
+output = 32_768

providers/abacus/models/moonshotai/Kimi-K2.7-Code.toml

@@ -0,0 +1,15 @@
+# Source: https://routellm.abacus.ai/v1/models
+# RouteLLM input_modalities: ['text', 'image', 'video']
+# RouteLLM output_modalities: ['text']
+# RouteLLM context_length: 262144
+# RouteLLM max_completion_tokens: 262144
+# RouteLLM input_token_rate: 0.00000095
+# RouteLLM output_token_rate: 0.000004
+# RouteLLM cached_input_token_rate: 0.00000019
+base_model = "moonshotai/kimi-k2.7-code"
+reasoning_options = []
+
+[cost]
+input = 0.95
+output = 4
+cache_read = 0.19

providers/abacus/models/moonshotai/Kimi-K3.toml

@@ -0,0 +1,15 @@
+# Source: https://routellm.abacus.ai/v1/models
+# RouteLLM input_modalities: ['text', 'image', 'video']
+# RouteLLM output_modalities: ['text']
+# RouteLLM context_length: 1048576
+# RouteLLM max_completion_tokens: 131072
+# RouteLLM input_token_rate: 0.000003
+# RouteLLM output_token_rate: 0.000015
+# RouteLLM cached_input_token_rate: 0.0000003
+base_model = "moonshotai/kimi-k3"
+reasoning_options = []
+
+[cost]
+input = 3
+output = 15
+cache_read = 0.3

providers/abacus/models/muse-spark-1.2.toml

@@ -0,0 +1,20 @@
+# Source: https://routellm.abacus.ai/v1/models
+# RouteLLM input_modalities: ['text', 'image', 'audio', 'video']
+# RouteLLM output_modalities: ['text']
+# RouteLLM context_length: 1048576
+# RouteLLM max_completion_tokens: 131072
+# RouteLLM input_token_rate: 0.00000125
+# RouteLLM output_token_rate: 0.00000425
+# RouteLLM cached_input_token_rate: 0.00000015
+# RouteLLM thinking: True
+base_model = "meta/muse-spark-1.2"
+reasoning_options = []
+
+[cost]
+input = 1.25
+output = 4.25
+cache_read = 0.15
+
+[modalities]
+input = ["text", "image", "audio", "video"]
+output = ["text"]

providers/abacus/models/qwen3.7-max.toml

@@ -0,0 +1,17 @@
+# Source: https://routellm.abacus.ai/v1/models
+# RouteLLM input_modalities: ['text']
+# RouteLLM output_modalities: ['text']
+# RouteLLM context_length: 1000000
+# RouteLLM max_completion_tokens: 64000
+# RouteLLM input_token_rate: 0.0000025
+# RouteLLM output_token_rate: 0.0000075
+# RouteLLM thinking: True
+base_model = "alibaba/qwen3.7-max"
+reasoning_options = []
+
+[cost]
+input = 2.5
+output = 7.5
+
+[limit]
+output = 64_000

providers/abacus/models/qwen3.8-max.toml

@@ -0,0 +1,18 @@
+# Source: https://routellm.abacus.ai/v1/models
+# RouteLLM input_modalities: ['text', 'image', 'video']
+# RouteLLM output_modalities: ['text']
+# RouteLLM context_length: 1000000
+# RouteLLM max_completion_tokens: 131072
+# RouteLLM input_token_rate: 0.000002
+# RouteLLM output_token_rate: 0.000006
+# RouteLLM thinking: True
+base_model = "alibaba/qwen3.8-max"
+reasoning_options = []
+
+[cost]
+input = 2
+output = 6
+
+[modalities]
+input = ["text", "image", "video"]
+output = ["text"]

providers/abacus/models/thinkingmachines/Inkling.toml

@@ -0,0 +1,24 @@
+# Source: https://routellm.abacus.ai/v1/models
+# RouteLLM input_modalities: ['text', 'image']
+# RouteLLM output_modalities: ['text']
+# RouteLLM context_length: 262144
+# RouteLLM max_completion_tokens: 131072
+# RouteLLM input_token_rate: 0.00000374
+# RouteLLM output_token_rate: 0.00000936
+# RouteLLM cached_input_token_rate: 0.000000748
+# RouteLLM thinking: True
+base_model = "thinkingmachines/inkling"
+reasoning_options = []
+
+[cost]
+input = 3.74
+output = 9.36
+cache_read = 0.748
+
+[limit]
+context = 262_144
+output = 131_072
+
+[modalities]
+input = ["text", "image"]
+output = ["text"]

providers/aiand/models/deepseek-ai/deepseek-v4-flash.toml

@@ -17,4 +17,3 @@ output = 0.25
 
 [limit]
 context = 1_048_576
-output = 384_000

providers/aiand/models/deepseek-ai/deepseek-v4-pro.toml

@@ -11,9 +11,8 @@ type = "effort"
 values = ["none", "minimal", "low", "medium", "high", "xhigh"]
 
 [cost]
-input = 1.00
-output = 2.50
+input = 1
+output = 2.5
 
 [limit]
 context = 1_048_576
-output = 384_000

providers/aiand/models/google/gemma-4-31b-it.toml

@@ -14,9 +14,8 @@ type = "effort"
 values = ["none", "minimal", "low", "medium", "high", "xhigh"]
 
 [cost]
-input = 0.20
-output = 0.50
+input = 0.2
+output = 0.5
 
 [modalities]
 input = ["text", "image", "video", "pdf"]
-output = ["text"]

providers/aiand/models/moonshotai/kimi-k2.7-code.toml

@@ -13,8 +13,7 @@ values = ["none", "minimal", "low", "medium", "high", "xhigh"]
 
 [cost]
 input = 0.75
-output = 3.50
+output = 3.5
 
 [modalities]
 input = ["text", "image", "pdf"]
-output = ["text"]

providers/aiand/models/moonshotai/kimi-k3.toml

@@ -6,19 +6,21 @@
 # text+image+video. PDF kept: sibling aiand Moonshot entries (kimi-k2.6,
 # kimi-k2.7-code) include pdf after catalog/probe evidence; aiand treats
 # PDF as a provider-level Files API modality.
-# reasoning_effort verified live: gateway schema accepts
-# none/minimal/low/medium/high/xhigh/max, but the K3 backend only accepts
-# none/low/high/max (minimal/medium/xhigh rejected). Invalid values rejected
-# with 400 (negative control).
+# reasoning_effort: K3 backend accepts only low/high/max per official docs
+# (https://platform.kimi.ai/docs/guide/kimi-k3-quickstart). The aiand gateway
+# schema lists none/minimal/low/medium/high/xhigh/max but K3 rejects
+# none/minimal/medium/xhigh with 400. aiand does not expose a separate off
+# toggle; K3 always reasons.
 base_model = "moonshotai/kimi-k3"
 
-reasoning_options = [{ type = "effort", values = ["none", "low", "high", "max"] }]
+[[reasoning_options]]
+type = "effort"
+values = ["low", "high", "max"]
 
 [cost]
-input = 3.00
-cache_read = 0.50
-output = 12.50
+input = 3
+output = 12.5
+cache_read = 0.5
 
 [modalities]
 input = ["text", "image", "pdf"]
-output = ["text"]

providers/aiand/models/motif-technologies/motif-3.toml

@@ -0,0 +1,23 @@
+name = "Motif 3"
+description = "Motif 3 is a large-scale, decoder-only Mixture-of-Experts (MoE) language model with 314 billion total parameters and 13.2 billion parameters activated per token."
+release_date = "2026-08-12"
+last_updated = "2026-08-12"
+attachment = false
+reasoning = true
+temperature = false
+tool_call = false
+structured_output = false
+open_weights = false
+reasoning_options = []
+
+[cost]
+input = 0.5
+output = 2
+
+[limit]
+context = 262_144
+output = 262_144
+
+[modalities]
+input = ["text"]
+output = ["text"]

providers/aiand/models/qwen/qwen3.6-27b.toml

@@ -1,6 +1,7 @@
-# Source: https://docs.aiand.com/models/catalog/ (USD list prices; accessed
-# 2026-07-24) and live probes against api.aiand.com on the same date.
-# Listed as free ("no charges, ideal for evals") in the catalog's Quick picks.
+# Source: GET https://api.aiand.com/v1/models (USD list prices; accessed
+# 2026-07-28). The catalog Quick-picks section still lists this model as free,
+# but the live endpoint reports input = $0.32 / output = $3.20 per 1M tokens,
+# so pricing is taken from the API.
 # ai&'s deployment rejects image input ("does not support image input") and
 # the catalog lists no vision/video/audio capability, so modalities are
 # overridden from the shared base model's text+image+video+audio to text-only,
@@ -9,16 +10,13 @@
 # values accepted; invalid values rejected with 400 (negative control).
 base_model = "alibaba/qwen3.6-27b"
 
-attachment = false
-
 [[reasoning_options]]
 type = "effort"
 values = ["none", "minimal", "low", "medium", "high", "xhigh"]
 
 [cost]
-input = 0
-output = 0
+input = 0.32
+output = 3.2
 
 [modalities]
-input = ["text"]
-output = ["text"]
+input = ["text", "image", "video", "pdf"]

providers/aiand/models/zai-org/glm-5.2.toml

@@ -11,9 +11,8 @@ type = "effort"
 values = ["none", "minimal", "low", "medium", "high", "xhigh"]
 
 [cost]
-input = 1.00
-output = 4.00
+input = 1
+output = 4
 
 [limit]
 context = 1_048_576
-output = 131_072

providers/aihubmix/models/claude-opus-5.toml

@@ -0,0 +1,15 @@
+# AIHubMix Anthropic-compatible /v1/messages: $.thinking.type = "disabled"|"adaptive" (toggle) and $.output_config.effort = "low"|"medium"|"high"|"xhigh"|"max"; verified live 2026-08-11. https://docs.aihubmix.com/cn/api-reference/anthropic-compatible/create-a-message
+base_model = "anthropic/claude-opus-5"
+structured_output = true
+reasoning_options = [{ type = "toggle" }, { type = "effort", values = ["low", "medium", "high", "xhigh", "max"] }]
+interleaved = true
+
+[cost]
+input = 5
+output = 25
+cache_read = 0.5
+cache_write = 6.25
+
+[modalities]
+input = ["text", "image"]
+output = ["text"]

providers/aihubmix/models/gemini-3.7-flash.toml

@@ -0,0 +1,10 @@
+base_model = "google/gemini-3.7-flash"
+
+# AIHubMix unified Chat: $.reasoning_effort = low|medium|high
+# https://docs.aihubmix.com/cn/api/unified-inference
+reasoning_options = [{ type = "effort", values = ["low", "medium", "high"] }]
+
+[cost]
+input = 0.75
+output = 3.75
+cache_read = 0.075