Add Crusoe provider (#3769)

* Add Crusoe provider

* Remove pricing; add Nemotron-3-Ultra-550B

* Address review: declare reasoning_options, theme-adaptive logo

- Add reasoning_options = [] to the 12 reasoning-model TOMLs: Crusoe's
  OpenAI-compatible endpoint documents no caller-side reasoning controls
  (docs.crusoecloud.com defers to the generic OpenAI API reference), so
  an empty declaration is correct per the validate schema.
- logo.svg: drop fixed width/height, use fill="currentColor" so the
  wordmark adapts to light/dark themes.

bun validate passes locally.

* Move reasoning_options rationale comments above first key

* Restore trailing newlines in reasoning-model TOMLs

* fix(crusoe): set reasoning config from live endpoint probe

Probed api.inference.crusoecloud.com on 2026-08-13 with reasoning_effort
low/medium/high/none/max plus tool-call interleaving checks per model.

- gpt-oss-120b: effort low/medium/high (reasoning length scales; none/max
  return 400), interleaved with tool calls
- GLM-5.2, Kimi-K2.6, Nemotron-3-Nano-Omni-Reasoning: toggle (effort
  "none" disables reasoning; low/medium/high inert), interleaved
- GLM-5.1: reasoning always on, no working caller-side control
- Reasoning arrives in the message field named "reasoning", so the
  boolean interleaved form is used
- Drop reasoning_options = [] from non-reasoning models
- Remove six models whose IDs drifted from the live /v1/models catalog
  or whose reasoning deployment is unverified; follow-up will re-add

* fix(crusoe): gemma-4-31b-it reasoning toggle

Base model has reasoning = true so reasoning_options is required by the
schema. Probe shows reasoning_effort acts as an enable/disable toggle on
this deployment (off by default, "none" disables, other values enable).

* feat(crusoe): add per-model pricing

Source: https://www.crusoe.ai/cloud/pricing (accessed 2026-08-13).
Input, output, and cached-read rates per million tokens for all eight
models. Nemotron Omni carries a separate audio input rate (0.50) via
cost.input_audio; its text/image/video input rate is 0.30.

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"

providers/crusoe/models/deepseek-ai/DeepSeek-V3-0324.toml

@@ -0,0 +1,7 @@
+base_model = "deepseek/deepseek-v3-0324"
+
+# Pricing: https://www.crusoe.ai/cloud/pricing (accessed 2026-08-13)
+[cost]
+input = 0.50
+output = 1.50
+cache_read = 0.25

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

@@ -0,0 +1,15 @@
+# Probed live endpoint 2026-08-13 (POST /v1/chat/completions, temperature 0):
+# reasoning is off by default (no reasoning field on a plain request);
+# setting reasoning_effort to any of low/medium/high enables it (constant
+# length, values inert relative to each other) and "none" disables it,
+# so this behaves as a toggle. Reasoning text arrives in the message field
+# named "reasoning" (not "reasoning_content"). A tool-call request returned
+# tool_calls without reasoning text, so no interleaved claim is made.
+base_model = "google/gemma-4-31b-it"
+reasoning_options = [{ type = "toggle" }]
+
+# Pricing: https://www.crusoe.ai/cloud/pricing (accessed 2026-08-13)
+[cost]
+input = 0.14
+output = 0.40
+cache_read = 0.14

providers/crusoe/models/meta-llama/Llama-3.3-70B-Instruct.toml

@@ -0,0 +1,7 @@
+base_model = "meta/llama-3.3-70b-instruct"
+
+# Pricing: https://www.crusoe.ai/cloud/pricing (accessed 2026-08-13)
+[cost]
+input = 0.25
+output = 0.75
+cache_read = 0.13

providers/crusoe/models/moonshotai/Kimi-K2.6.toml

@@ -0,0 +1,15 @@
+# Probed live endpoint 2026-08-13 (POST /v1/chat/completions, temperature 0):
+# reasoning is on by default; reasoning_effort "none" disables it (0 chars);
+# low/medium/high are accepted but do not materially change reasoning
+# length, so this is encoded as a toggle (matches deepinfra's Kimi-K2.6
+# entry). Reasoning text arrives interleaved with tool_calls, in the
+# message field named "reasoning" (not "reasoning_content").
+base_model = "moonshotai/kimi-k2.6"
+reasoning_options = [{ type = "toggle" }]
+interleaved = true
+
+# Pricing: https://www.crusoe.ai/cloud/pricing (accessed 2026-08-13)
+[cost]
+input = 0.70
+output = 3.50
+cache_read = 0.35

providers/crusoe/models/nvidia/Nemotron-3-Nano-Omni-Reasoning-30B-A3B.toml

@@ -0,0 +1,17 @@
+# Probed live endpoint 2026-08-13 (POST /v1/chat/completions, temperature 0):
+# reasoning is on by default; reasoning_effort "none" disables it (0 chars);
+# low/medium/high are accepted but inert; "max" returns HTTP 400. Encoded
+# as a toggle. Reasoning text arrives interleaved with tool_calls, in the
+# message field named "reasoning" (not "reasoning_content").
+base_model = "nvidia/nemotron-3-nano-omni-30b-a3b-reasoning"
+reasoning_options = [{ type = "toggle" }]
+interleaved = true
+
+# Pricing: https://www.crusoe.ai/cloud/pricing (accessed 2026-08-13).
+# Text/image/video input is 0.30; audio input is billed at 0.50
+# (cached-audio pricing exists on the page but has no schema field).
+[cost]
+input = 0.30
+output = 1.83
+cache_read = 0.30
+input_audio = 0.50

providers/crusoe/models/openai/gpt-oss-120b.toml

@@ -0,0 +1,14 @@
+# Probed live endpoint 2026-08-13 (POST /v1/chat/completions, temperature 0):
+# reasoning_effort low/medium/high accepted and scale reasoning length
+# (26/58/174 chars on the same prompt); "none" and "max" return HTTP 400.
+# Reasoning text arrives interleaved with tool_calls, in the message field
+# named "reasoning" (not "reasoning_content"), hence the boolean form below.
+base_model = "openai/gpt-oss-120b"
+reasoning_options = [{ type = "effort", values = ["low", "medium", "high"] }]
+interleaved = true
+
+# Pricing: https://www.crusoe.ai/cloud/pricing (accessed 2026-08-13)
+[cost]
+input = 0.05
+output = 0.20
+cache_read = 0.05

providers/crusoe/models/zai/GLM-5.1.toml

@@ -0,0 +1,15 @@
+# Probed live endpoint 2026-08-13 (POST /v1/chat/completions, temperature 0):
+# reasoning is always on. reasoning_effort low/medium/high/none/max are all
+# accepted but inert (identical reasoning length on every value; "none"
+# does not disable). No working caller-side control was found, so no
+# options are advertised. Reasoning text arrives interleaved with
+# tool_calls, in the message field named "reasoning".
+base_model = "zhipuai/glm-5.1"
+reasoning_options = []
+interleaved = true
+
+# Pricing: https://www.crusoe.ai/cloud/pricing (accessed 2026-08-13)
+[cost]
+input = 1.20
+output = 4.40
+cache_read = 0.25

providers/crusoe/models/zai/GLM-5.2.toml

@@ -0,0 +1,16 @@
+# Probed live endpoint 2026-08-13 (POST /v1/chat/completions, temperature 0):
+# reasoning is on by default; reasoning_effort "none" disables it
+# (0 reasoning tokens); low/medium/high are accepted but do not scale
+# reasoning length, so this is encoded as a toggle (matches the zai lab
+# GLM entries). usage.completion_tokens_details.reasoning_tokens is
+# reported. Reasoning text arrives interleaved with tool_calls, in the
+# message field named "reasoning" (not "reasoning_content").
+base_model = "zhipuai/glm-5.2"
+reasoning_options = [{ type = "toggle" }]
+interleaved = true
+
+# Pricing: https://www.crusoe.ai/cloud/pricing (accessed 2026-08-13)
+[cost]
+input = 1.40
+output = 4.40
+cache_read = 0.26

providers/crusoe/provider.toml

@@ -0,0 +1,5 @@
+name = "Crusoe"
+env = ["CRUSOE_API_KEY"]
+npm = "@ai-sdk/openai-compatible"
+api = "https://api.inference.crusoecloud.com/v1"
+doc = "https://docs.crusoecloud.com/managed-inference/overview"