chore(sync): update OpenRouter model catalog (#5316)

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

.github/workflows/sync-models.yml

@@ -95,6 +95,7 @@ jobs:
         run: bun validate
 
       - name: Report changes
+        id: report
         env:
           GH_TOKEN: ${{ steps.committer.outputs.token }}
           BRANCH: automation/sync-models-${{ matrix.provider }}
@@ -119,9 +120,20 @@ jobs:
           git checkout -B "$BRANCH"
           git add models providers
           git commit -m "$TITLE"
-          git push --force-with-lease origin "$BRANCH"
+          bun sync:auto-merge HEAD^ HEAD
+          safe="$(sed -n 's/^safe=//p' "$GITHUB_OUTPUT" | tail -1)"
 
           pr_number="$(gh pr list --head "$BRANCH" --base dev --json number --jq '.[0].number')"
+          if [ "$safe" != "true" ] && [ -n "$pr_number" ]; then
+            gh pr merge "$pr_number" --disable-auto || true
+            if [ "$(gh pr view "$pr_number" --json autoMergeRequest --jq '.autoMergeRequest == null')" != "true" ]; then
+              echo "Failed to disable auto-merge for unsafe sync PR #$pr_number."
+              exit 1
+            fi
+          fi
+
+          git push --force-with-lease origin "$BRANCH"
+
           if [ -n "$pr_number" ]; then
             gh pr edit "$pr_number" --title "$TITLE" --body-file .sync/model-sync-report.md
             for label in "${labels[@]}"; do
@@ -129,4 +141,12 @@ jobs:
             done
           else
             gh pr create --base dev --head "$BRANCH" --title "$TITLE" --body-file .sync/model-sync-report.md "${label_args[@]}"
+            pr_number="$(gh pr list --head "$BRANCH" --base dev --json number --jq '.[0].number')"
+          fi
+
+          if [ "$safe" = "true" ]; then
+            gh pr merge "$pr_number" --auto --squash
+          elif [ "$(gh pr view "$pr_number" --json autoMergeRequest --jq '.autoMergeRequest == null')" != "true" ]; then
+            echo "Unsafe sync PR #$pr_number still has auto-merge enabled."
+            exit 1
           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-0.5b.toml

@@ -0,0 +1,23 @@
+name = "Qwen2.5-Coder-0.5B"
+description = "Tiny open Qwen code model for lightweight completion and on-device coding"
+family = "qwen"
+release_date = "2024-11-12"
+last_updated = "2024-11-12"
+attachment = false
+reasoning = false
+temperature = true
+tool_call = false
+open_weights = true
+license = "Apache 2.0"
+
+[limit]
+context = 32_768
+output = 8_192
+
+[modalities]
+input = ["text"]
+output = ["text"]
+
+[[weights]]
+label = "Hugging Face"
+url = "https://huggingface.co/Qwen/Qwen2.5-Coder-0.5B"

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/qwen3-235b-a22b-instruct-2507.toml

@@ -0,0 +1,23 @@
+name = "Qwen3 235B-A22B Instruct 2507"
+description = "Updated large open Qwen3 MoE instruct model for multilingual chat, coding, and tool use"
+family = "qwen"
+release_date = "2025-07-21"
+last_updated = "2025-07-21"
+attachment = false
+reasoning = false
+temperature = true
+tool_call = true
+open_weights = true
+license = "Apache 2.0"
+
+[limit]
+context = 262_144
+output = 16_384
+
+[modalities]
+input = ["text"]
+output = ["text"]
+
+[[weights]]
+label = "Hugging Face"
+url = "https://huggingface.co/Qwen/Qwen3-235B-A22B-Instruct-2507"

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-vl-235b-a22b-instruct.toml

@@ -0,0 +1,24 @@
+name = "Qwen3 VL 235B A22B Instruct"
+description = "Qwen vision-language instruct model for visual reasoning, documents, and agent tasks"
+family = "qwen"
+release_date = "2025-09-23"
+last_updated = "2025-09-23"
+attachment = true
+reasoning = false
+temperature = true
+tool_call = true
+structured_output = true
+knowledge = "2025-03-31"
+open_weights = true
+
+[limit]
+context = 131_072
+output = 32_768
+
+[modalities]
+input = ["text", "image"]
+output = ["text"]
+
+[[weights]]
+label = "Hugging Face"
+url = "https://huggingface.co/Qwen/Qwen3-VL-235B-A22B-Instruct"

models/alibaba/qwen3-vl-235b-a22b-thinking.toml

@@ -0,0 +1,24 @@
+name = "Qwen3 VL 235B A22B Thinking"
+description = "Qwen vision-language thinking model for visual reasoning, documents, and agent tasks"
+family = "qwen"
+release_date = "2025-09-23"
+last_updated = "2025-09-23"
+attachment = true
+reasoning = true
+temperature = true
+tool_call = true
+structured_output = true
+knowledge = "2025-03-31"
+open_weights = true
+
+[limit]
+context = 131_072
+output = 32_768
+
+[modalities]
+input = ["text", "image"]
+output = ["text"]
+
+[[weights]]
+label = "Hugging Face"
+url = "https://huggingface.co/Qwen/Qwen3-VL-235B-A22B-Thinking"

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/qwen3.7-flash.toml

@@ -0,0 +1,20 @@
+name = "Qwen3.7 Flash"
+description = "Lightweight multimodal Qwen model for high-throughput text, image, and video tasks"
+family = "qwen"
+release_date = "2026-07-15"
+last_updated = "2026-07-15"
+attachment = true
+reasoning = true
+temperature = true
+tool_call = true
+structured_output = true
+open_weights = false
+
+[limit]
+context = 1_000_000
+input = 991_000
+output = 65_536
+
+[modalities]
+input = ["text", "image", "video"]
+output = ["text"]

models/alibaba/qwen3.8-2.4t-a95b.toml

@@ -0,0 +1,33 @@
+# Sources (accessed 2026-08-16):
+# https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B
+# https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B/raw/main/README.md
+# https://qwen.ai/blog?id=qwen3.8
+# https://openrouter.ai/qwen/qwen3.8-2.4t-a95b
+# Open-weight twin of Qwen3.8 Max: text-only, thinking always on,
+# reasoning_effort low|medium|xhigh (default xhigh). Native context 262K,
+# extensible to ~1.01M. Distinct from closed multimodal qwen3.8-max.
+
+name = "Qwen3.8 2.4T A95B"
+description = "Open-weight sparse MoE (2.4T total, 95B active), the open-weight twin of Qwen3.8 Max for coding, research, complex reasoning, and agentic workflows"
+family = "qwen"
+release_date = "2026-08-12"
+last_updated = "2026-08-12"
+attachment = false
+reasoning = true
+temperature = true
+tool_call = true
+structured_output = true
+open_weights = true
+license = "qwen3.8-max"
+
+[limit]
+context = 262_144
+output = 131_072
+
+[modalities]
+input = ["text"]
+output = ["text"]
+
+[[weights]]
+label = "Hugging Face"
+url = "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B"

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-preview.toml

@@ -28,3 +28,131 @@ output = 131_072
 [modalities]
 input = ["text", "image", "video"]
 output = ["text"]
+
+[[benchmarks]]
+name = "Terminal-Bench"
+score = 86.6
+metric = "accuracy"
+variant = "xhigh"
+version = "2.1"
+source = "https://www.alibabacloud.com/blog/qwen3-8-max-a-new-bar-for-coding-and-cowork_603421"
+date = "2026-08-03"
+
+[[benchmarks]]
+name = "SWE-Bench Pro"
+score = 67.7
+metric = "resolve rate"
+variant = "xhigh"
+harness = "Claude Code"
+source = "https://www.alibabacloud.com/blog/qwen3-8-max-a-new-bar-for-coding-and-cowork_603421"
+date = "2026-08-03"
+
+[[benchmarks]]
+name = "DeepSWE"
+score = 56.6
+metric = "resolve rate"
+variant = "xhigh"
+harness = "Claude Code"
+version = "1.1"
+source = "https://www.alibabacloud.com/blog/qwen3-8-max-a-new-bar-for-coding-and-cowork_603421"
+date = "2026-08-03"
+
+[[benchmarks]]
+name = "NL2Repo"
+score = 55.9
+metric = "resolve rate"
+variant = "xhigh"
+harness = "Claude Code"
+source = "https://www.alibabacloud.com/blog/qwen3-8-max-a-new-bar-for-coding-and-cowork_603421"
+date = "2026-08-03"
+
+[[benchmarks]]
+name = "FrontierSWE"
+score = 73.5
+metric = "dominance score"
+variant = "xhigh"
+harness = "Claude Code"
+source = "https://www.alibabacloud.com/blog/qwen3-8-max-a-new-bar-for-coding-and-cowork_603421"
+date = "2026-08-03"
+
+[[benchmarks]]
+name = "MLS-Bench-Lite"
+score = 41.0
+metric = "score"
+variant = "xhigh"
+harness = "Claude Code"
+source = "https://www.alibabacloud.com/blog/qwen3-8-max-a-new-bar-for-coding-and-cowork_603421"
+date = "2026-08-03"
+
+[[benchmarks]]
+name = "AutomationBench"
+score = 27.3
+metric = "pass@1"
+variant = "xhigh"
+dataset = "600-task public subset"
+source = "https://www.alibabacloud.com/blog/qwen3-8-max-a-new-bar-for-coding-and-cowork_603421"
+date = "2026-08-03"
+
+[[benchmarks]]
+name = "Toolathlon Verified"
+score = 72.5
+metric = "pass@1"
+variant = "xhigh"
+source = "https://www.alibabacloud.com/blog/qwen3-8-max-a-new-bar-for-coding-and-cowork_603421"
+date = "2026-08-03"
+
+[[benchmarks]]
+name = "WideSearch"
+score = 81.9
+metric = "F1"
+variant = "xhigh"
+source = "https://www.alibabacloud.com/blog/qwen3-8-max-a-new-bar-for-coding-and-cowork_603421"
+date = "2026-08-03"
+
+[[benchmarks]]
+name = "Humanity's Last Exam"
+score = 56.2
+metric = "accuracy"
+variant = "xhigh, with tools"
+source = "https://www.alibabacloud.com/blog/qwen3-8-max-a-new-bar-for-coding-and-cowork_603421"
+date = "2026-08-03"
+
+[[benchmarks]]
+name = "GPQA Diamond"
+score = 92.6
+metric = "accuracy"
+variant = "xhigh"
+source = "https://www.alibabacloud.com/blog/qwen3-8-max-a-new-bar-for-coding-and-cowork_603421"
+date = "2026-08-03"
+
+[[benchmarks]]
+name = "Humanity's Last Exam"
+score = 43.6
+metric = "accuracy"
+variant = "xhigh, no tools"
+source = "https://www.alibabacloud.com/blog/qwen3-8-max-a-new-bar-for-coding-and-cowork_603421"
+date = "2026-08-03"
+
+[[benchmarks]]
+name = "IFBench"
+score = 82.8
+metric = "score"
+variant = "xhigh"
+source = "https://www.alibabacloud.com/blog/qwen3-8-max-a-new-bar-for-coding-and-cowork_603421"
+date = "2026-08-03"
+
+[[benchmarks]]
+name = "OSWorld-Verified"
+score = 86.1
+metric = "success rate"
+variant = "xhigh"
+source = "https://www.alibabacloud.com/blog/qwen3-8-max-a-new-bar-for-coding-and-cowork_603421"
+date = "2026-08-03"
+
+[[benchmarks]]
+name = "MMMU Pro"
+score = 82.3
+metric = "accuracy"
+variant = "xhigh"
+source = "https://www.alibabacloud.com/blog/qwen3-8-max-a-new-bar-for-coding-and-cowork_603421"
+date = "2026-08-03"

models/alibaba/qwen3.8-max.toml

@@ -0,0 +1,38 @@
+# 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
+# https://docs.qwencloud.com/developer-guides/getting-started/text-generation-models
+# https://docs.qwencloud.com/developer-guides/text-generation/thinking
+# 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"
+family = "qwen"
+release_date = "2026-08-03"
+last_updated = "2026-08-03"
+attachment = true
+reasoning = true
+temperature = true
+tool_call = true
+open_weights = false
+
+[limit]
+context = 1_000_000
+output = 131_072
+
+[modalities]
+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/anthropic/claude-opus-5.toml

@@ -17,3 +17,156 @@ output = 128_000
 [modalities]
 input = ["text", "image", "pdf"]
 output = ["text"]
+
+[[benchmarks]]
+name = "SWE-Bench Verified"
+score = 96.0
+metric = "resolved"
+source = "https://www.anthropic.com/claude-opus-5-system-card"
+date = "2026-07-24"
+
+[[benchmarks]]
+name = "SWE-Bench Pro"
+score = 79.2
+metric = "resolve rate"
+source = "https://www.anthropic.com/claude-opus-5-system-card"
+date = "2026-07-24"
+
+[[benchmarks]]
+name = "SWE-Bench Multilingual"
+score = 89.5
+metric = "resolve rate"
+source = "https://www.anthropic.com/claude-opus-5-system-card"
+date = "2026-07-24"
+
+[[benchmarks]]
+name = "SWE-Bench Multimodal"
+score = 59.4
+metric = "resolve rate"
+source = "https://www.anthropic.com/claude-opus-5-system-card"
+date = "2026-07-24"
+
+[[benchmarks]]
+name = "DeepSWE"
+score = 68.8
+metric = "resolve rate"
+version = "1.1"
+source = "https://www.anthropic.com/claude-opus-5-system-card"
+date = "2026-07-24"
+
+[[benchmarks]]
+name = "FrontierCode"
+score = 53.4
+metric = "mean@5"
+variant = "medium effort"
+dataset = "Main"
+version = "1.1"
+source = "https://www.anthropic.com/claude-opus-5-system-card"
+date = "2026-07-24"
+
+[[benchmarks]]
+name = "Frontier-Bench"
+score = 43.3
+metric = "mean reward"
+variant = "max effort"
+harness = "mini-SWE-agent"
+version = "v0.1"
+source = "https://www.anthropic.com/claude-opus-5-system-card"
+date = "2026-07-24"
+
+[[benchmarks]]
+name = "BrowseComp"
+score = 90.8
+metric = "accuracy"
+variant = "single agent"
+source = "https://www.anthropic.com/claude-opus-5-system-card"
+date = "2026-07-24"
+
+[[benchmarks]]
+name = "Humanity's Last Exam"
+score = 56.3
+metric = "accuracy"
+variant = "no tools"
+source = "https://www.anthropic.com/claude-opus-5-system-card"
+date = "2026-07-24"
+
+[[benchmarks]]
+name = "Humanity's Last Exam"
+score = 64.7
+metric = "accuracy"
+variant = "with tools"
+source = "https://www.anthropic.com/claude-opus-5-system-card"
+date = "2026-07-24"
+
+[[benchmarks]]
+name = "DeepSearchQA"
+score = 95.0
+metric = "F1"
+variant = "max effort"
+source = "https://www.anthropic.com/news/claude-opus-5"
+date = "2026-07-24"
+
+[[benchmarks]]
+name = "OSWorld"
+score = 70.6
+metric = "success rate"
+version = "2.0"
+source = "https://www.anthropic.com/claude-opus-5-system-card"
+date = "2026-07-24"
+
+[[benchmarks]]
+name = "GDPval-AA"
+score = 1861
+metric = "Elo"
+variant = "max effort"
+version = "v2"
+source = "https://www.anthropic.com/claude-opus-5-system-card"
+date = "2026-07-24"
+
+[[benchmarks]]
+name = "AA-Briefcase"
+score = 1720
+metric = "Elo"
+variant = "max effort"
+source = "https://www.anthropic.com/claude-opus-5-system-card"
+date = "2026-07-24"
+
+[[benchmarks]]
+name = "AutomationBench"
+score = 26.0
+metric = "success rate"
+variant = "max effort"
+source = "https://www.anthropic.com/claude-opus-5-system-card"
+date = "2026-07-24"
+
+[[benchmarks]]
+name = "ARC-AGI-1"
+score = 97.5
+metric = "accuracy"
+variant = "max effort"
+source = "https://www.anthropic.com/claude-opus-5-system-card"
+date = "2026-07-24"
+
+[[benchmarks]]
+name = "ARC-AGI-2"
+score = 90.4
+metric = "accuracy"
+variant = "max effort"
+source = "https://www.anthropic.com/claude-opus-5-system-card"
+date = "2026-07-24"
+
+[[benchmarks]]
+name = "ARC-AGI-3"
+score = 30.2
+metric = "RHAE"
+variant = "high effort"
+source = "https://www.anthropic.com/claude-opus-5-system-card"
+date = "2026-07-24"
+
+[[benchmarks]]
+name = "HealthBench Professional"
+score = 59.8
+metric = "score"
+variant = "max effort"
+source = "https://www.anthropic.com/claude-opus-5-system-card"
+date = "2026-07-24"

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-v3.1.toml

@@ -0,0 +1,23 @@
+name = "DeepSeek-V3.1"
+description = "Hybrid-reasoning DeepSeek model with thinking and non-thinking modes"
+family = "deepseek"
+release_date = "2025-08-21"
+last_updated = "2025-08-21"
+attachment = false
+reasoning = true
+temperature = true
+tool_call = true
+open_weights = true
+license = "MIT License"
+
+[limit]
+context = 131_072
+output = 8_192
+
+[modalities]
+input = ["text"]
+output = ["text"]
+
+[[weights]]
+label = "Hugging Face"
+url = "https://huggingface.co/deepseek-ai/DeepSeek-V3.1"

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-v3.toml

@@ -0,0 +1,23 @@
+name = "DeepSeek-V3"
+description = "Open DeepSeek MoE chat model for coding, math, and general reasoning"
+family = "deepseek"
+release_date = "2024-12-26"
+last_updated = "2024-12-26"
+attachment = false
+reasoning = false
+temperature = true
+tool_call = true
+open_weights = true
+license = "DeepSeek Model License"
+
+[limit]
+context = 131_072
+output = 8_192
+
+[modalities]
+input = ["text"]
+output = ["text"]
+
+[[weights]]
+label = "Hugging Face"
+url = "https://huggingface.co/deepseek-ai/DeepSeek-V3"

models/deepseek/deepseek-v4-flash-0731.toml

@@ -0,0 +1,105 @@
+name = "DeepSeek V4 Flash 0731"
+description = "Official DeepSeek V4 Flash release with enhanced agentic capabilities and integrated DSpark speculative decoding"
+family = "deepseek-flash"
+release_date = "2026-07-31"
+last_updated = "2026-07-31"
+attachment = false
+reasoning = true
+temperature = true
+tool_call = true
+structured_output = true
+knowledge = "2025-05"
+open_weights = true
+license = "MIT"
+
+[limit]
+context = 1_000_000
+output = 384_000
+
+[modalities]
+input = ["text"]
+output = ["text"]
+
+[[weights]]
+label = "Hugging Face"
+url = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731"
+
+[[benchmarks]]
+name = "Terminal-Bench"
+score = 82.7
+metric = "pass@1"
+variant = "max"
+version = "2.1"
+source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731"
+
+[[benchmarks]]
+name = "NL2Repo"
+score = 54.2
+metric = "resolve rate"
+variant = "max effort"
+harness = "DeepSeek Harness minimal mode"
+source = "https://api-docs.deepseek.com/updates/"
+date = "2026-07-31"
+
+[[benchmarks]]
+name = "CyberGym"
+score = 76.7
+metric = "score"
+variant = "max effort"
+harness = "DeepSeek Harness minimal mode"
+source = "https://api-docs.deepseek.com/updates/"
+date = "2026-07-31"
+
+[[benchmarks]]
+name = "DeepSWE"
+score = 54.4
+metric = "resolve rate"
+variant = "max effort"
+harness = "DeepSeek Harness minimal mode"
+source = "https://api-docs.deepseek.com/updates/"
+date = "2026-07-31"
+
+[[benchmarks]]
+name = "Toolathlon-Verified"
+score = 70.3
+metric = "score"
+variant = "max effort"
+harness = "DeepSeek Harness minimal mode"
+source = "https://api-docs.deepseek.com/updates/"
+date = "2026-07-31"
+
+[[benchmarks]]
+name = "Agents' Last Exam"
+score = 25.2
+metric = "score"
+variant = "max effort"
+harness = "DeepSeek Harness minimal mode"
+source = "https://api-docs.deepseek.com/updates/"
+date = "2026-07-31"
+
+[[benchmarks]]
+name = "AutomationBench"
+score = 25.1
+metric = "success rate"
+variant = "max effort"
+dataset = "public"
+source = "https://api-docs.deepseek.com/updates/"
+date = "2026-07-31"
+
+[[benchmarks]]
+name = "DSBench-FullStack"
+score = 68.7
+metric = "score"
+variant = "max effort"
+dataset = "internal"
+source = "https://api-docs.deepseek.com/updates/"
+date = "2026-07-31"
+
+[[benchmarks]]
+name = "DSBench-Hard"
+score = 59.6
+metric = "score"
+variant = "max effort"
+dataset = "internal"
+source = "https://api-docs.deepseek.com/updates/"
+date = "2026-07-31"

models/deepseek/deepseek-v4-flash-vision-exp.toml

@@ -0,0 +1,19 @@
+name = "DeepSeek V4 Flash Vision Exp"
+description = "Experimental multimodal DeepSeek V4 Flash model for image understanding, coding, and agentic work"
+family = "deepseek-flash"
+release_date = "2026-08-21"
+last_updated = "2026-08-21"
+attachment = true
+reasoning = true
+temperature = true
+tool_call = true
+structured_output = true
+open_weights = false
+
+[limit]
+context = 1_000_000
+output = 384_000
+
+[modalities]
+input = ["text", "image"]
+output = ["text"]

models/deepseek/deepseek-v4-flash.toml

@@ -1,12 +1,8 @@
-# DeepSeek-V4-Flash-0731 official API (public beta): same architecture as preview, re-post-trained.
-# https://api-docs.deepseek.com/updates (2026-07-31)
-# https://api-docs.deepseek.com/quick_start/pricing/
-# https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731
 name = "DeepSeek V4 Flash"
 description = "Fast DeepSeek V4 lane for economical reasoning, coding, and long-context work"
 family = "deepseek-flash"
 release_date = "2026-04-24"
-last_updated = "2026-07-31"
+last_updated = "2026-04-24"
 attachment = false
 reasoning = true
 temperature = true
@@ -25,29 +21,10 @@ output = ["text"]
 
 [[weights]]
 label = "Hugging Face"
-url = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731"
+url = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash"
 
 [[benchmarks]]
-name = "Terminal-Bench"
-score = 82.7
-metric = "pass@1"
-version = "2.1"
-source = "https://api-docs.deepseek.com/updates"
-
-[[benchmarks]]
-name = "Toolathlon"
-score = 70.3
-metric = "verified"
-source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731"
-
-[[benchmarks]]
-name = "NL2Repo"
-score = 54.2
-metric = "score"
-source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731"
-
-[[benchmarks]]
-name = "DeepSWE"
-score = 54.4
-metric = "score"
-source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731"
+name = "SWE-Bench Verified"
+score = 79
+metric = "resolved"
+source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash"

models/deepseek/deepseek-v4-pro-0423.toml

@@ -0,0 +1,22 @@
+# https://ofox.ai/models/deepseek/deepseek-v4-pro-0423
+# https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro
+name = "DeepSeek V4 Pro 0423"
+description = "DeepSeek V4 Pro initial snapshot with million-token context and support for thinking and non-thinking modes"
+family = "deepseek-thinking"
+release_date = "2026-04-23"
+last_updated = "2026-04-23"
+attachment = false
+reasoning = true
+temperature = true
+tool_call = true
+structured_output = true
+knowledge = "2025-05"
+open_weights = true
+
+[limit]
+context = 1_000_000
+output = 384_000
+
+[modalities]
+input = ["text"]
+output = ["text"]

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

@@ -0,0 +1,25 @@
+# https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813
+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 = "2026-08-22"
+attachment = false
+reasoning = true
+temperature = true
+tool_call = true
+structured_output = true
+open_weights = true
+license = "MIT"
+
+[limit]
+context = 1_000_000
+output = 384_000
+
+[modalities]
+input = ["text"]
+output = ["text"]
+
+[[weights]]
+label = "Hugging Face"
+url = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813"

models/google/gemini-3.5-flash-lite.toml

@@ -17,4 +17,76 @@ output = 65_536
 
 [modalities]
 input = ["text", "image", "video", "audio", "pdf"]
-output = ["text"]
+output = ["text"]
+
+[[benchmarks]]
+name = "SWE-Bench Pro"
+score = 54.2
+metric = "resolve rate"
+source = "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/"
+date = "2026-07-21"
+
+[[benchmarks]]
+name = "Terminal-Bench"
+score = 54.0
+metric = "accuracy"
+harness = "Terminus 2"
+version = "2.1"
+source = "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/"
+date = "2026-07-21"
+
+[[benchmarks]]
+name = "MLE-Bench"
+score = 39.2
+metric = "average position score"
+source = "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/"
+date = "2026-07-21"
+
+[[benchmarks]]
+name = "GDPval-AA"
+score = 1140
+metric = "Elo"
+version = "v2"
+source = "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/"
+date = "2026-07-21"
+
+[[benchmarks]]
+name = "OSWorld-Verified"
+score = 74.0
+metric = "success rate"
+source = "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/"
+date = "2026-07-21"
+
+[[benchmarks]]
+name = "CharXiv Reasoning"
+score = 74.5
+metric = "accuracy"
+variant = "no tools"
+source = "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/"
+date = "2026-07-21"
+
+[[benchmarks]]
+name = "CharXiv Reasoning"
+score = 76.5
+metric = "accuracy"
+variant = "with tools"
+source = "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/"
+date = "2026-07-21"
+
+[[benchmarks]]
+name = "GDM-MRCR"
+score = 72.2
+metric = "accuracy"
+variant = "128k average, 8-needle"
+version = "v2"
+source = "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/"
+date = "2026-07-21"
+
+[[benchmarks]]
+name = "GDM-MRCR"
+score = 21.3
+metric = "accuracy"
+variant = "1M pointwise, 8-needle"
+version = "v2"
+source = "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/"
+date = "2026-07-21"

models/google/gemini-3.6-flash.toml

@@ -17,4 +17,87 @@ output = 65_536
 
 [modalities]
 input = ["text", "image", "video", "audio", "pdf"]
-output = ["text"]
+output = ["text"]
+
+[[benchmarks]]
+name = "SWE-Bench Pro"
+score = 58.7
+metric = "resolve rate"
+harness = "Antigravity"
+source = "https://deepmind.google/models/evals-methodology/gemini-3-6-flash/"
+date = "2026-07-21"
+
+[[benchmarks]]
+name = "DeepSWE"
+score = 49.0
+metric = "resolve rate"
+variant = "high reasoning"
+version = "1.1"
+source = "https://deepmind.google/models/evals-methodology/gemini-3-6-flash/"
+date = "2026-07-21"
+
+[[benchmarks]]
+name = "Terminal-Bench"
+score = 78.0
+metric = "accuracy"
+harness = "Terminus 2"
+version = "2.1"
+source = "https://deepmind.google/models/evals-methodology/gemini-3-6-flash/"
+date = "2026-07-21"
+
+[[benchmarks]]
+name = "MLE-Bench"
+score = 63.9
+metric = "average position score"
+dataset = "Partial 30"
+source = "https://deepmind.google/models/evals-methodology/gemini-3-6-flash/"
+date = "2026-07-21"
+
+[[benchmarks]]
+name = "GDPval-AA"
+score = 1421
+metric = "Elo"
+version = "v2"
+source = "https://deepmind.google/models/evals-methodology/gemini-3-6-flash/"
+date = "2026-07-21"
+
+[[benchmarks]]
+name = "OSWorld-Verified"
+score = 83.0
+metric = "success rate"
+source = "https://deepmind.google/models/evals-methodology/gemini-3-6-flash/"
+date = "2026-07-21"
+
+[[benchmarks]]
+name = "CharXiv Reasoning"
+score = 85.2
+metric = "accuracy"
+variant = "no tools"
+source = "https://deepmind.google/models/evals-methodology/gemini-3-6-flash/"
+date = "2026-07-21"
+
+[[benchmarks]]
+name = "CharXiv Reasoning"
+score = 89.4
+metric = "accuracy"
+variant = "with tools"
+source = "https://deepmind.google/models/evals-methodology/gemini-3-6-flash/"
+date = "2026-07-21"
+
+[[benchmarks]]
+name = "GDM-MRCR"
+score = 91.8
+metric = "accuracy"
+variant = "128k average, 8-needle"
+version = "v2"
+source = "https://deepmind.google/models/evals-methodology/gemini-3-6-flash/"
+date = "2026-07-21"
+
+[[benchmarks]]
+name = "GDM-MRCR"
+score = 54.0
+metric = "accuracy"
+variant = "1M pointwise, 8-needle"
+version = "v2"
+source = "https://deepmind.google/models/evals-methodology/gemini-3-6-flash/"
+date = "2026-07-21"

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/google/gemini-flash-latest.toml

@@ -1,15 +1,15 @@
-# Tracks the current Gemini Flash release (gemini-3.5-flash).
+# Tracks the current Gemini Flash release (gemini-3.7-flash).
 name = "Gemini Flash Latest"
-description = "Fast Gemini model balancing multimodal reasoning, tool use, and cost"
+description = "High-efficiency Gemini model for agentic workflows, coding, and multimodal reasoning"
 family = "gemini-flash"
-release_date = "2026-05-19"
-last_updated = "2026-05-19"
+release_date = "2026-08-13"
+last_updated = "2026-08-13"
 attachment = true
 reasoning = true
 temperature = true
 tool_call = true
 structured_output = true
-knowledge = "2025-01"
+knowledge = "2026-03"
 open_weights = false
 
 [limit]

models/google/gemini-flash-lite-latest.toml

@@ -1,15 +1,15 @@
-# Tracks the current Gemini Flash-Lite release (gemini-3.1-flash-lite).
+# Tracks the current Gemini Flash-Lite release (gemini-3.5-flash-lite).
 name = "Gemini Flash-Lite Latest"
-description = "Low-latency Gemini model for high-volume multimodal and agent workloads"
+description = "Fast Gemini model balancing multimodal reasoning, tool use, and cost"
 family = "gemini-flash-lite"
-release_date = "2026-05-07"
-last_updated = "2026-05-07"
+release_date = "2026-07-21"
+last_updated = "2026-07-21"
 attachment = true
 reasoning = true
 temperature = true
 tool_call = true
 structured_output = true
-knowledge = "2025-01"
+knowledge = "2026-03"
 open_weights = false
 
 [limit]

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-3.2-1b.toml

@@ -0,0 +1,24 @@
+name = "Llama-3.2-1B"
+description = "Compact open Llama base model for lightweight and on-device use"
+family = "llama"
+release_date = "2024-09-25"
+last_updated = "2024-09-25"
+attachment = false
+reasoning = false
+temperature = true
+tool_call = false
+knowledge = "2023-12"
+open_weights = true
+license = "Llama 3.2 Community License"
+
+[limit]
+context = 131_072
+output = 8_192
+
+[modalities]
+input = ["text"]
+output = ["text"]
+
+[[weights]]
+label = "Hugging Face"
+url = "https://huggingface.co/meta-llama/Llama-3.2-1B"

models/meta/llama-3.2-3b.toml

@@ -0,0 +1,24 @@
+name = "Llama-3.2-3B"
+description = "Small open Llama base model for lightweight text generation and self-hosting"
+family = "llama"
+release_date = "2024-09-25"
+last_updated = "2024-09-25"
+attachment = false
+reasoning = false
+temperature = true
+tool_call = false
+knowledge = "2023-12"
+open_weights = true
+license = "Llama 3.2 Community License"
+
+[limit]
+context = 131_072
+output = 8_192
+
+[modalities]
+input = ["text"]
+output = ["text"]
+
+[[weights]]
+label = "Hugging Face"
+url = "https://huggingface.co/meta-llama/Llama-3.2-3B"

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/codestral-22b-v0.1.toml

@@ -0,0 +1,23 @@
+name = "Codestral-22B-v0.1"
+description = "Open Mistral code model for fill-in-the-middle and 80+ programming languages"
+family = "codestral"
+release_date = "2024-05-29"
+last_updated = "2024-05-29"
+attachment = false
+reasoning = false
+temperature = true
+tool_call = false
+open_weights = true
+license = "Mistral AI Non-Production License"
+
+[limit]
+context = 32_768
+output = 8_192
+
+[modalities]
+input = ["text"]
+output = ["text"]
+
+[[weights]]
+label = "Hugging Face"
+url = "https://huggingface.co/mistralai/Codestral-22B-v0.1"

models/mistral/magistral-small-2506.toml

@@ -0,0 +1,23 @@
+name = "Magistral Small"
+description = "Open Mistral reasoning model for transparent step-by-step problem solving"
+family = "magistral"
+release_date = "2025-06-10"
+last_updated = "2025-06-10"
+attachment = false
+reasoning = true
+temperature = true
+tool_call = true
+open_weights = true
+license = "Apache 2.0"
+
+[limit]
+context = 131_072
+output = 8_192
+
+[modalities]
+input = ["text"]
+output = ["text"]
+
+[[weights]]
+label = "Hugging Face"
+url = "https://huggingface.co/mistralai/Magistral-Small-2506"

models/mistral/ministral-8b-instruct-2410.toml

@@ -0,0 +1,23 @@
+name = "Ministral 8B Instruct"
+description = "Efficient open Mistral edge model for on-device chat and function calling"
+family = "ministral"
+release_date = "2024-10-16"
+last_updated = "2024-10-16"
+attachment = false
+reasoning = false
+temperature = true
+tool_call = true
+open_weights = true
+license = "Mistral Research License"
+
+[limit]
+context = 131_072
+output = 8_192
+
+[modalities]
+input = ["text"]
+output = ["text"]
+
+[[weights]]
+label = "Hugging Face"
+url = "https://huggingface.co/mistralai/Ministral-8B-Instruct-2410"

models/mistral/mistral-medium-2604.toml

@@ -27,3 +27,11 @@ name = "SWE-Bench Verified"
 score = 77.6
 metric = "resolved"
 source = "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B"
+
+[[benchmarks]]
+name = "τ³-Telecom"
+score = 91.4
+metric = "accuracy"
+variant = "public preview"
+source = "https://mistral.ai/news/vibe-remote-agents-mistral-medium-3-5/"
+date = "2026-05-22"

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/mistral/voxtral-small-latest.toml

@@ -0,0 +1,24 @@
+# Sources (accessed 2026-08-16):
+#   https://docs.mistral.ai/models/model-cards/voxtral-small-25-07
+#   https://mistral.ai/news/voxtral/
+# Field values mirror Mistral's own first-party host entry in this repo
+# (providers/mistral/models/voxtral-small-latest.toml); host-scoped keys
+# (cost, status) are intentionally left to the provider files.
+name = "Voxtral Small (latest)"
+description = "Instruct model with native audio input for speech understanding and tool use"
+family = "voxtral"
+release_date = "2025-07-15"
+last_updated = "2025-07-15"
+attachment = true
+reasoning = false
+temperature = true
+tool_call = true
+open_weights = true
+
+[limit]
+context = 32_000
+output = 32_000
+
+[modalities]
+input = ["text", "audio"]
+output = ["text"]

models/moonshotai/kimi-k3.toml

@@ -17,3 +17,119 @@ output = 131_072
 [modalities]
 input = ["text", "image", "video"]
 output = ["text"]
+
+[[benchmarks]]
+name = "DeepSWE"
+score = 67.5
+metric = "resolve rate"
+variant = "max effort"
+harness = "Kimi Code"
+version = "1.1"
+source = "https://www.kimi.com/blog/kimi-k3"
+date = "2026-07-16"
+
+[[benchmarks]]
+name = "Terminal-Bench"
+score = 88.3
+metric = "accuracy"
+variant = "max effort"
+harness = "Kimi Code"
+version = "2.1"
+source = "https://www.kimi.com/blog/kimi-k3"
+date = "2026-07-16"
+
+[[benchmarks]]
+name = "FrontierSWE"
+score = 81.2
+metric = "dominance score"
+variant = "max effort"
+harness = "Kimi Code"
+source = "https://www.kimi.com/blog/kimi-k3"
+date = "2026-07-16"
+
+[[benchmarks]]
+name = "Program Bench"
+score = 77.8
+metric = "score"
+variant = "max effort"
+harness = "Kimi Code"
+source = "https://www.kimi.com/blog/kimi-k3"
+date = "2026-07-16"
+
+[[benchmarks]]
+name = "SWE Marathon"
+score = 42.0
+metric = "resolve rate"
+variant = "max effort"
+harness = "Claude Code"
+version = "1.1"
+source = "https://www.kimi.com/blog/kimi-k3"
+date = "2026-07-16"
+
+[[benchmarks]]
+name = "GDPval-AA"
+score = 1668
+metric = "Elo"
+variant = "max effort"
+version = "v2"
+source = "https://www.kimi.com/blog/kimi-k3"
+date = "2026-07-16"
+
+[[benchmarks]]
+name = "AA-Briefcase"
+score = 1548
+metric = "Elo"
+variant = "max effort"
+source = "https://www.kimi.com/blog/kimi-k3"
+date = "2026-07-16"
+
+[[benchmarks]]
+name = "AutomationBench"
+score = 30.8
+metric = "success rate"
+variant = "max effort"
+dataset = "600-task public subset"
+source = "https://www.kimi.com/blog/kimi-k3"
+date = "2026-07-16"
+
+[[benchmarks]]
+name = "JobBench"
+score = 52.9
+metric = "score"
+variant = "max effort"
+source = "https://www.kimi.com/blog/kimi-k3"
+date = "2026-07-16"
+
+[[benchmarks]]
+name = "SpreadsheetBench"
+score = 34.8
+metric = "score"
+variant = "max effort"
+harness = "Claude Code"
+version = "2"
+source = "https://www.kimi.com/blog/kimi-k3"
+date = "2026-07-16"
+
+[[benchmarks]]
+name = "BrowseComp"
+score = 91.2
+metric = "accuracy"
+variant = "max effort, context compaction"
+source = "https://www.kimi.com/blog/kimi-k3"
+date = "2026-07-16"
+
+[[benchmarks]]
+name = "CharXiv Reasoning"
+score = 91.3
+metric = "accuracy"
+variant = "max effort, with tools"
+source = "https://www.kimi.com/blog/kimi-k3"
+date = "2026-07-16"
+
+[[benchmarks]]
+name = "ZeroBench"
+score = 41.0
+metric = "pass@5"
+variant = "max effort, with tools"
+source = "https://www.kimi.com/blog/kimi-k3"
+date = "2026-07-16"

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

@@ -1,25 +1,20 @@
-name = "GPT-5.3-Codex"
+name = "GPT-5.3 Codex Spark"
 description = "Coding-optimized GPT model for repository edits, reviews, and agentic software work"
-family = "gpt-codex"
-release_date = "2026-02-24"
-last_updated = "2026-02-24"
+family = "gpt-codex-spark"
+release_date = "2026-02-05"
+last_updated = "2026-02-05"
 attachment = true
 reasoning = true
-reasoning_options = [{ type = "effort", values = ["none", "low", "medium", "high", "max"] }, { type = "budget_tokens" }]
 temperature = false
 knowledge = "2025-08-31"
 tool_call = true
 structured_output = true
 open_weights = false
 
-[cost]
-input = 1.75
-output = 14.00
-cache_read = 0.175
-
 [limit]
-context = 400_000
-output = 128_000
+context = 128_000
+input = 100_000
+output = 32_000
 
 [modalities]
 input = ["text", "image", "pdf"]

models/openai/gpt-5.4-mini.toml

@@ -19,3 +19,135 @@ output = 128_000
 [modalities]
 input = ["text", "image"]
 output = ["text"]
+
+[[benchmarks]]
+name = "SWE-Bench Pro"
+score = 54.4
+metric = "resolve rate"
+variant = "reasoning effort xhigh"
+source = "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/"
+date = "2026-03-17"
+
+[[benchmarks]]
+name = "Terminal-Bench"
+score = 60.0
+metric = "accuracy"
+variant = "reasoning effort xhigh"
+version = "2.0"
+source = "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/"
+date = "2026-03-17"
+
+[[benchmarks]]
+name = "MCP Atlas"
+score = 57.7
+metric = "score"
+variant = "reasoning effort xhigh"
+source = "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/"
+date = "2026-03-17"
+
+[[benchmarks]]
+name = "Toolathlon"
+score = 42.9
+metric = "score"
+variant = "reasoning effort xhigh"
+source = "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/"
+date = "2026-03-17"
+
+[[benchmarks]]
+name = "τ²-Bench Telecom"
+score = 93.4
+metric = "accuracy"
+variant = "reasoning effort xhigh"
+source = "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/"
+date = "2026-03-17"
+
+[[benchmarks]]
+name = "GPQA Diamond"
+score = 88.0
+metric = "accuracy"
+variant = "reasoning effort xhigh"
+source = "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/"
+date = "2026-03-17"
+
+[[benchmarks]]
+name = "Humanity's Last Exam"
+score = 41.5
+metric = "accuracy"
+variant = "with tools"
+source = "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/"
+date = "2026-03-17"
+
+[[benchmarks]]
+name = "Humanity's Last Exam"
+score = 28.2
+metric = "accuracy"
+variant = "without tools"
+source = "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/"
+date = "2026-03-17"
+
+[[benchmarks]]
+name = "OSWorld-Verified"
+score = 72.1
+metric = "success rate"
+variant = "reasoning effort xhigh"
+source = "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/"
+date = "2026-03-17"
+
+[[benchmarks]]
+name = "MMMU Pro"
+score = 78.0
+metric = "accuracy"
+variant = "with Python"
+source = "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/"
+date = "2026-03-17"
+
+[[benchmarks]]
+name = "MMMU Pro"
+score = 76.6
+metric = "accuracy"
+variant = "without tools"
+source = "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/"
+date = "2026-03-17"
+
+[[benchmarks]]
+name = "OmniDocBench"
+score = 0.1263
+metric = "overall edit distance"
+variant = "reasoning effort none"
+version = "1.5"
+source = "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/"
+date = "2026-03-17"
+
+[[benchmarks]]
+name = "OpenAI MRCR"
+score = 47.7
+metric = "accuracy"
+variant = "8-needle, 64K-128K"
+version = "v2"
+source = "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/"
+date = "2026-03-17"
+
+[[benchmarks]]
+name = "OpenAI MRCR"
+score = 33.6
+metric = "accuracy"
+variant = "8-needle, 128K-256K"
+version = "v2"
+source = "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/"
+date = "2026-03-17"
+
+[[benchmarks]]
+name = "Graphwalks"
+score = 76.3
+metric = "accuracy"
+variant = "BFS, 0-128K"
+source = "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/"
+date = "2026-03-17"
+
+[[benchmarks]]
+name = "Graphwalks"
+score = 71.5
+metric = "accuracy"
+variant = "parents, 0-128K"
+source = "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/"
+date = "2026-03-17"

models/openai/gpt-5.4-nano.toml

@@ -19,3 +19,135 @@ output = 128_000
 [modalities]
 input = ["text", "image"]
 output = ["text"]
+
+[[benchmarks]]
+name = "SWE-Bench Pro"
+score = 52.4
+metric = "resolve rate"
+variant = "reasoning effort xhigh"
+source = "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/"
+date = "2026-03-17"
+
+[[benchmarks]]
+name = "Terminal-Bench"
+score = 46.3
+metric = "accuracy"
+variant = "reasoning effort xhigh"
+version = "2.0"
+source = "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/"
+date = "2026-03-17"
+
+[[benchmarks]]
+name = "MCP Atlas"
+score = 56.1
+metric = "score"
+variant = "reasoning effort xhigh"
+source = "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/"
+date = "2026-03-17"
+
+[[benchmarks]]
+name = "Toolathlon"
+score = 35.5
+metric = "score"
+variant = "reasoning effort xhigh"
+source = "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/"
+date = "2026-03-17"
+
+[[benchmarks]]
+name = "τ²-Bench Telecom"
+score = 92.5
+metric = "accuracy"
+variant = "reasoning effort xhigh"
+source = "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/"
+date = "2026-03-17"
+
+[[benchmarks]]
+name = "GPQA Diamond"
+score = 82.8
+metric = "accuracy"
+variant = "reasoning effort xhigh"
+source = "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/"
+date = "2026-03-17"
+
+[[benchmarks]]
+name = "Humanity's Last Exam"
+score = 37.7
+metric = "accuracy"
+variant = "with tools"
+source = "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/"
+date = "2026-03-17"
+
+[[benchmarks]]
+name = "Humanity's Last Exam"
+score = 24.3
+metric = "accuracy"
+variant = "without tools"
+source = "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/"
+date = "2026-03-17"
+
+[[benchmarks]]
+name = "OSWorld-Verified"
+score = 39.0
+metric = "success rate"
+variant = "reasoning effort xhigh"
+source = "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/"
+date = "2026-03-17"
+
+[[benchmarks]]
+name = "MMMU Pro"
+score = 69.5
+metric = "accuracy"
+variant = "with Python"
+source = "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/"
+date = "2026-03-17"
+
+[[benchmarks]]
+name = "MMMU Pro"
+score = 66.1
+metric = "accuracy"
+variant = "without tools"
+source = "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/"
+date = "2026-03-17"
+
+[[benchmarks]]
+name = "OmniDocBench"
+score = 0.2419
+metric = "overall edit distance"
+variant = "reasoning effort none"
+version = "1.5"
+source = "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/"
+date = "2026-03-17"
+
+[[benchmarks]]
+name = "OpenAI MRCR"
+score = 44.2
+metric = "accuracy"
+variant = "8-needle, 64K-128K"
+version = "v2"
+source = "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/"
+date = "2026-03-17"
+
+[[benchmarks]]
+name = "OpenAI MRCR"
+score = 33.1
+metric = "accuracy"
+variant = "8-needle, 128K-256K"
+version = "v2"
+source = "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/"
+date = "2026-03-17"
+
+[[benchmarks]]
+name = "Graphwalks"
+score = 73.4
+metric = "accuracy"
+variant = "BFS, 0-128K"
+source = "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/"
+date = "2026-03-17"
+
+[[benchmarks]]
+name = "Graphwalks"
+score = 50.8
+metric = "accuracy"
+variant = "parents, 0-128K"
+source = "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/"
+date = "2026-03-17"

models/perplexity/sonar-deep-research.toml

@@ -0,0 +1,25 @@
+# Sources (accessed 2026-08-16):
+#   https://docs.perplexity.ai/docs/sonar/models/sonar-deep-research
+#   https://docs.perplexity.ai/api-reference/sonar-post
+# Field values mirror Perplexity's own first-party host entry in this repo
+# (providers/perplexity/models/sonar-deep-research.toml); host-scoped keys
+# (cost, reasoning_options) are intentionally left to the provider files.
+name = "Sonar Deep Research"
+description = "Sonar search model for autonomous research and citation-backed long-form reports"
+family = "sonar"
+release_date = "2025-02-01"
+last_updated = "2025-09-01"
+attachment = false
+reasoning = true
+temperature = false
+tool_call = false
+knowledge = "2025-01"
+open_weights = false
+
+[limit]
+context = 128_000
+output = 32_768
+
+[modalities]
+input = ["text"]
+output = ["text"]

models/sakana/sakana-namazu.toml

@@ -0,0 +1,59 @@
+name = "Sakana Namazu"
+description = "Japanese-specialized reasoning model based on Kimi K2.6 and tuned for Japanese language, culture, and business workflows"
+family = "sakana-namazu"
+release_date = "2026-08-03"
+last_updated = "2026-08-03"
+attachment = true
+reasoning = true
+temperature = true
+tool_call = true
+structured_output = true
+open_weights = false
+
+[limit]
+context = 262_144
+output = 65_536
+
+[modalities]
+input = ["text", "image", "pdf"]
+output = ["text"]
+
+[[links]]
+label = "Official product page"
+url = "https://sakana.ai/namazu/"
+type = "announcement"
+
+[[links]]
+label = "Official model documentation"
+url = "https://console.sakana.ai/models?model=sakana-namazu"
+type = "docs"
+
+[[benchmarks]]
+name = "AIME26"
+score = 96.67
+source = "https://console.sakana.ai/models?model=sakana-namazu"
+
+[[benchmarks]]
+name = "MMLU-Pro"
+score = 90.33
+source = "https://console.sakana.ai/models?model=sakana-namazu"
+
+[[benchmarks]]
+name = "LiveCodeBench v6"
+score = 90.33
+source = "https://console.sakana.ai/models?model=sakana-namazu"
+
+[[benchmarks]]
+name = "JFBench"
+score = 37.40
+source = "https://console.sakana.ai/models?model=sakana-namazu"
+
+[[benchmarks]]
+name = "Translation"
+score = 52.20
+source = "https://console.sakana.ai/models?model=sakana-namazu"
+
+[[benchmarks]]
+name = "FairPoliticsQA"
+score = 56.30
+source = "https://console.sakana.ai/models?model=sakana-namazu"

models/sdaia/allam-2-7b.toml

@@ -0,0 +1,21 @@
+name = "ALLaM-2-7b"
+description = "ALLaM-2-7b instruction tuned model by SDAIA"
+release_date = "2025-01-23"
+last_updated = "2025-01-23"
+attachment = false
+reasoning = false
+temperature = true
+tool_call = false
+open_weights = true
+
+[limit]
+context = 4096
+output = 4096
+
+[modalities]
+input = ["text"]
+output = ["text"]
+
+[[weights]]
+label = "Hugging Face"
+url = "https://huggingface.co/ALLaM-AI/ALLaM-2.0-7B-Instruct"

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/swiss-ai/apertus-8b.toml

@@ -0,0 +1,33 @@
+# Sources (accessed 2026-08-16):
+#   https://huggingface.co/swiss-ai/Apertus-8B-Instruct-2509
+#   https://arxiv.org/abs/2509.14233
+# The model card states "Apertus by default supports a context length up to 65,536 tokens",
+# Apache-2.0 licensing, and tool use support. Sibling entry: models/swiss-ai/apertus-70b.toml.
+name = "Apertus 8B"
+description = "Fully open 8B multilingual LLM supporting 1800+ languages with 65K context. Trained on 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-8B-Instruct-2509"
+
+[[links]]
+label = "Paper"
+url = "https://arxiv.org/abs/2509.14233"
+type = "paper"

models/trendyol/asure-12b.toml

@@ -0,0 +1,30 @@
+# Sources:
+# https://huggingface.co/Trendyol/Trendyol-LLM-Asure-12B
+# https://huggingface.co/api/models/Trendyol/Trendyol-LLM-Asure-12B (createdAt, license, base_model)
+# https://huggingface.co/Trendyol/Trendyol-LLM-Asure-12B/raw/main/config.json (max_position_embeddings)
+# `reasoning` and `tool_call` are not stated on the model card; both were
+# measured against a host serving these weights (llmtr.com, 2026-08-16):
+# a request carrying `tools` returns no tool_calls, and no reasoning output
+# is produced.
+name = "Trendyol Asure 12B"
+description = "Turkish-language multimodal instruct model built on Gemma 3 12B for e-commerce text, chat, and image-text tasks"
+family = "gemma"
+release_date = "2026-02-19"
+last_updated = "2026-02-20"
+attachment = true
+reasoning = false
+temperature = true
+tool_call = false
+open_weights = true
+license = "Gemma"
+
+[limit]
+context = 131_072
+
+[modalities]
+input = ["text", "image"]
+output = ["text"]
+
+[[weights]]
+label = "Hugging Face"
+url = "https://huggingface.co/Trendyol/Trendyol-LLM-Asure-12B"

models/upstage/solar-pro2.toml

@@ -0,0 +1,25 @@
+# Sources (accessed 2026-08-16):
+#   https://developers.upstage.ai/docs/apis/chat
+#   https://developers.upstage.ai/docs/capabilities/generate/reasoning
+# Field values mirror Upstage's own first-party host entry in this repo
+# (providers/upstage/models/solar-pro2.toml); host-scoped keys (cost,
+# reasoning_options) are intentionally left to the provider files.
+name = "Solar Pro 2"
+description = "Flagship model for demanding analysis, coding, and production agent workflows"
+family = "solar-pro"
+release_date = "2025-05-20"
+last_updated = "2025-05-20"
+attachment = false
+reasoning = true
+temperature = true
+knowledge = "2025-03"
+tool_call = true
+open_weights = false
+
+[limit]
+context = 65_536
+output = 8_192
+
+[modalities]
+input = ["text"]
+output = ["text"]

models/upstage/solar-pro3.toml

@@ -0,0 +1,25 @@
+# Sources (accessed 2026-08-16):
+#   https://developers.upstage.ai/docs/apis/chat
+#   https://developers.upstage.ai/docs/capabilities/generate/reasoning
+# Field values mirror Upstage's own first-party host entry in this repo
+# (providers/upstage/models/solar-pro3.toml); host-scoped keys (cost,
+# reasoning_options) are intentionally left to the provider files.
+name = "Solar Pro 3"
+description = "Flagship model for demanding analysis, coding, and production agent workflows"
+family = "solar-pro"
+release_date = "2026-01"
+last_updated = "2026-01"
+attachment = false
+reasoning = true
+temperature = true
+knowledge = "2025-03"
+tool_call = true
+open_weights = false
+
+[limit]
+context = 131_072
+output = 8_192
+
+[modalities]
+input = ["text"]
+output = ["text"]

models/upstage/solar-pro4.toml

@@ -0,0 +1,26 @@
+# Sources (accessed 2026-08-16):
+#   https://developers.upstage.ai/docs/apis/chat
+#   https://developers.upstage.ai/docs/capabilities/generate/reasoning
+# Field values mirror Upstage's own first-party host entry in this repo
+# (providers/upstage/models/solar-pro4.toml); host-scoped keys (cost,
+# reasoning_options) are intentionally left to the provider files.
+name = "Solar Pro 4"
+description = "Upstage's flagship model, specialized for agentic use"
+family = "solar-pro"
+release_date = "2026-08-06"
+last_updated = "2026-08-06"
+attachment = false
+reasoning = true
+temperature = true
+knowledge = "2026-02"
+tool_call = true
+structured_output = true
+open_weights = false
+
+[limit]
+context = 524_288
+output = 131_072
+
+[modalities]
+input = ["text"]
+output = ["text"]

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"
@@ -56,3 +56,10 @@ harness = "mini-swe-agent"
 version = "1.1"
 source = "https://x.ai/news/grok-4-5"
 date = "2026-07-08"
+
+[[benchmarks]]
+name = "SWE Marathon"
+score = 29.0
+metric = "pass@1"
+source = "https://x.ai/news/grok-4-5"
+date = "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/xai/grok-imagine-image-2.0.toml

@@ -0,0 +1,26 @@
+# Sources:
+# - https://docs.x.ai/docs/models
+# - https://docs.x.ai/developers/models/grok-imagine-image-2.0
+# - https://docs.x.ai/docs/guides/image-generation
+# - https://x.ai/news/grok-imagine-image-2
+# Pricing: $0.04 per image (not token-based; no [cost] authored)
+# Release: 2026-08-07 (GA as Quality Mode; API model id grok-imagine-image-2.0)
+
+name = "Grok Imagine Image 2.0"
+description = "Image model for prompt-driven generation, editing, and visual design workflows"
+family = "grok"
+release_date = "2026-08-07"
+last_updated = "2026-08-07"
+attachment = true
+reasoning = false
+temperature = false
+tool_call = false
+open_weights = false
+
+[limit]
+context = 8_000
+output = 0
+
+[modalities]
+input = ["text", "image"]
+output = ["image"]

models/zhipuai/glm-4.6v-flash.toml

@@ -0,0 +1,25 @@
+# Sources (accessed 2026-08-19):
+# - https://z.ai/blog/glm-4.6v
+# - https://huggingface.co/zai-org/GLM-4.6V-Flash
+name = "GLM-4.6V-Flash"
+description = "Lightweight GLM vision model for visual reasoning, documents, and multimodal agents"
+family = "glm"
+release_date = "2025-12-08"
+last_updated = "2025-12-08"
+attachment = true
+reasoning = true
+temperature = true
+tool_call = true
+open_weights = true
+
+[limit]
+context = 128_000
+output = 32_768
+
+[modalities]
+input = ["text", "image", "video"]
+output = ["text"]
+
+[[weights]]
+label = "Hugging Face"
+url = "https://huggingface.co/zai-org/GLM-4.6V-Flash"

models/zhipuai/glm-5.2.toml

@@ -44,3 +44,124 @@ score = 74.4
 metric = "dominance"
 source = "https://z.ai/blog/glm-5.2"
 date = "2026-06-16"
+
+[[benchmarks]]
+name = "Humanity's Last Exam"
+score = 40.5
+metric = "accuracy"
+dataset = "text-only subset"
+source = "https://z.ai/blog/glm-5.2"
+date = "2026-06-16"
+
+[[benchmarks]]
+name = "Humanity's Last Exam"
+score = 54.7
+metric = "accuracy"
+variant = "with tools"
+dataset = "text-only subset"
+source = "https://z.ai/blog/glm-5.2"
+date = "2026-06-16"
+
+[[benchmarks]]
+name = "CritPt"
+score = 20.9
+metric = "accuracy"
+source = "https://z.ai/blog/glm-5.2"
+date = "2026-06-16"
+
+[[benchmarks]]
+name = "AIME"
+score = 99.2
+metric = "accuracy"
+version = "2026"
+source = "https://z.ai/blog/glm-5.2"
+date = "2026-06-16"
+
+[[benchmarks]]
+name = "HMMT"
+score = 94.4
+metric = "accuracy"
+version = "November 2025"
+source = "https://z.ai/blog/glm-5.2"
+date = "2026-06-16"
+
+[[benchmarks]]
+name = "HMMT"
+score = 92.5
+metric = "accuracy"
+version = "February 2026"
+source = "https://z.ai/blog/glm-5.2"
+date = "2026-06-16"
+
+[[benchmarks]]
+name = "IMOAnswerBench"
+score = 91.0
+metric = "accuracy"
+source = "https://z.ai/blog/glm-5.2"
+date = "2026-06-16"
+
+[[benchmarks]]
+name = "GPQA Diamond"
+score = 91.2
+metric = "accuracy"
+source = "https://z.ai/blog/glm-5.2"
+date = "2026-06-16"
+
+[[benchmarks]]
+name = "NL2Repo"
+score = 48.9
+metric = "resolve rate"
+source = "https://z.ai/blog/glm-5.2"
+date = "2026-06-16"
+
+[[benchmarks]]
+name = "DeepSWE"
+score = 46.2
+metric = "resolve rate"
+source = "https://z.ai/blog/glm-5.2"
+date = "2026-06-16"
+
+[[benchmarks]]
+name = "Program Bench"
+score = 63.7
+metric = "score"
+source = "https://z.ai/blog/glm-5.2"
+date = "2026-06-16"
+
+[[benchmarks]]
+name = "Terminal-Bench"
+score = 81.0
+metric = "success rate"
+harness = "Terminus 2"
+version = "2.1"
+source = "https://z.ai/blog/glm-5.2"
+date = "2026-06-16"
+
+[[benchmarks]]
+name = "PostTrainBench"
+score = 34.3
+metric = "score"
+source = "https://z.ai/blog/glm-5.2"
+date = "2026-06-16"
+
+[[benchmarks]]
+name = "SWE Marathon"
+score = 13.0
+metric = "resolve rate"
+source = "https://z.ai/blog/glm-5.2"
+date = "2026-06-16"
+
+[[benchmarks]]
+name = "MCP Atlas"
+score = 76.8
+metric = "score"
+dataset = "public subset"
+source = "https://z.ai/blog/glm-5.2"
+date = "2026-06-16"
+
+[[benchmarks]]
+name = "Tool-Decathlon"
+score = 48.2
+metric = "score"
+source = "https://z.ai/blog/glm-5.2"
+date = "2026-06-16"

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"]

package.json

@@ -28,6 +28,8 @@
     "huggingface:sync": "bun ./packages/core/script/sync-models.ts huggingface",
     "kilo:sync": "bun ./packages/core/script/sync-models.ts kilo",
     "llmgateway:sync": "bun ./packages/core/script/sync-models.ts llmgateway",
+    "llmgateway-providers:sync": "bun ./packages/core/script/sync-models.ts llmgateway-providers",
+    "requesty:sync": "bun ./packages/core/script/sync-models.ts requesty",
     "merge-gateway:sync": "bun ./packages/core/script/sync-models.ts merge-gateway",
     "nano-gpt:sync": "bun ./packages/core/script/sync-models.ts nano-gpt",
     "venice:sync": "bun ./packages/core/script/sync-models.ts venice",
@@ -37,7 +39,8 @@
     "digitalocean:sync": "bun ./packages/core/script/sync-models.ts digitalocean",
     "ambient:sync": "bun ./packages/core/script/sync-models.ts ambient",
     "models:sync": "bun ./packages/core/script/sync-models.ts",
-    "sync:models": "bun ./packages/core/script/sync-models.ts"
+    "sync:models": "bun ./packages/core/script/sync-models.ts",
+    "sync:auto-merge": "bun ./packages/core/script/check-sync-auto-merge.ts"
   },
   "dependencies": {
     "@cloudflare/workers-types": "^4.20260424.1",

packages/core/script/check-sync-auto-merge.ts

@@ -0,0 +1,31 @@
+import { appendFile } from "node:fs/promises";
+
+import { classifyAutoMerge, parseNameStatus } from "../src/sync/auto-merge.js";
+
+const base = process.argv[2] ?? "HEAD^";
+const head = process.argv[3] ?? "HEAD";
+const diff = Bun.spawnSync(["git", "diff", "--name-status", "--no-renames", base, head], {
+  stdout: "pipe",
+  stderr: "inherit",
+});
+
+if (diff.exitCode !== 0) process.exit(diff.exitCode ?? 1);
+
+const loadPrevious = async (path: string) => {
+  const file = Bun.spawnSync(["git", "show", `${base}:${path}`], {
+    stdout: "pipe",
+    stderr: "inherit",
+  });
+  if (file.exitCode !== 0) throw new Error(`Failed to read ${path} at ${base}`);
+  return file.stdout.toString();
+};
+
+const decision = await classifyAutoMerge(parseNameStatus(diff.stdout.toString()), undefined, loadPrevious);
+const summary = decision.safe
+  ? `Safe to auto-merge: ${decision.created} created, ${decision.updated} updated, ${decision.deleted} deleted.`
+  : `Manual review required: ${decision.reasons.join("; ")}.`;
+
+console.log(summary);
+if (process.env.GITHUB_OUTPUT) {
+  await appendFile(process.env.GITHUB_OUTPUT, `safe=${decision.safe}\nsummary=${summary}\n`);
+}

packages/core/src/family.ts

@@ -30,6 +30,7 @@ export const ModelFamilyValues = [
   "claude-sonnet",
   "claude-opus",
   "claude-fable",
+  "claude-mythos",
 
   // Gemini style
   "gemini",
@@ -51,6 +52,7 @@ export const ModelFamilyValues = [
 
   // Meta Muse
   "muse",
+  "muse-free",
 
   // Alibaba Qwen
   "qwen",
@@ -58,6 +60,7 @@ export const ModelFamilyValues = [
   "qwen3.6",
   "qwen3.7-plus",
   "qwen3.7-max",
+  "qwen3.8-max",
   "qwen-free",
 
   // DeepReinforce
@@ -364,6 +367,9 @@ export const ModelFamilyValues = [
   // Conductor
   "fugu",
 
+  // Sakana Namazu
+  "sakana-namazu",
+
   // V0
   "v0",
 

packages/core/src/generate.ts

@@ -89,6 +89,11 @@ async function generateProviders(
     }
 
     const modelsPath = path.join(directory, providerID, "models");
+    if (!existsSync(modelsPath)) {
+      throw new Error(`Provider "${providerID}" has no models`, {
+        cause: { providerPath },
+      });
+    }
     for await (const modelPath of new Bun.Glob("**/*.toml").scan({
       cwd: modelsPath,
       absolute: true,
@@ -124,6 +129,11 @@ async function generateProviders(
       }
       provider.data.models[modelID] = normalizeModelCost(model.data);
     }
+    if (Object.keys(provider.data.models).length === 0) {
+      throw new Error(`Provider "${providerID}" has no models`, {
+        cause: { providerPath },
+      });
+    }
     result[providerID] = provider.data;
   }