✗ WRONG

pydantic-ai 2.61.0

request.max-tokens-ollama-openai · ollama-openai

Requirements

Id	Level	Status	Evidence
limit-field	MUST	CONTRADICTED	request[0].json_body.max_tokens=absent; required=16

What the provider sent

Finish: length
Raw finish: length
Text: Here are three moons of Jupiter: 1. **Callisto** (
Tool calls: []
Usage: { "alternatives": {}, "cached_input": 0, "input": 18, "output": 16, "raw": { "completion_tokens": 16, "prompt_tokens": 18, "prompt_tokens_details": { "cached_tokens": 0 }, "total_tokens": 34 }, "reasoning": null, "total": 34 }
Stream completed: null
Error: null

Wire excerpt

status: 200
content-type: application/json
body (json):
{
  "id": "chatcmpl-390",
  "object": "chat.completion",
  "created": 1791831669,
  "model": "qwen3.5:0.8b",
  "system_fingerprint": "fp_ollama",
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "content": "Here are three moons of Jupiter:\n\n1.  **Callisto** ("
      },
      "finish_reason": "length"
    }
  ],
  "usage": {
    "prompt_tokens": 18,
    "prompt_tokens_details": {
      "cached_tokens": 0
    },
    "completion_tokens": 16,
    "total_tokens": 34
  }
}

What the library reported

Outcome

returned

Finish

{
  "raw": "length",
  "source": "ModelResponse.finish_reason",
  "surface": "primary",
  "value": "length"
}

Text

Value: Here are three moons of Jupiter: 1. **Callisto** (
Surface: primary

Tool calls

[]

Usage

{
  "cached_input": 0,
  "input": 18,
  "output": 16,
  "reasoning": 0,
  "surface": {
    "cached_input": "primary",
    "input": "primary",
    "output": "primary",
    "reasoning": "primary",
    "total": "primary"
  },
  "total": 34
}

Stream

{
  "chunks_seen": null,
  "completed": null,
  "used": false
}

Error

null

Unsupported reason

null

Full observation JSON

{
  "elapsed_ms": 223.8046284765005,
  "error": null,
  "finish": {
    "raw": "length",
    "source": "ModelResponse.finish_reason",
    "surface": "primary",
    "value": "length"
  },
  "format": "wiretruth.observation/v1",
  "library": {
    "api": "Agent.run",
    "language": "python",
    "name": "pydantic-ai",
    "version": "2.61.0"
  },
  "notes": [],
  "outcome": "returned",
  "stream": {
    "chunks_seen": null,
    "completed": null,
    "used": false
  },
  "structured": {
    "attempted": false,
    "returned_as_success": false,
    "value": null
  },
  "text": {
    "surface": "primary",
    "value": "Here are three moons of Jupiter:\n\n1.  **Callisto** ("
  },
  "tool_calls": [],
  "unsupported_reason": null,
  "usage": {
    "cached_input": 0,
    "input": 18,
    "output": 16,
    "reasoning": 0,
    "surface": {
      "cached_input": "primary",
      "input": "primary",
      "output": "primary",
      "reasoning": "primary",
      "total": "primary"
    },
    "total": 34
  }
}

Fixture

recorded ollama-openai/request.max-tokens-ollama-openai

Provenance	Value
Recorded at	2026-10-12T19:01:09Z
Requested model	qwen3.5:0.8b
Returned model	qwen3.5:0.8b
CI run	https://github.com/kartsan03/wiretruth/actions/runs/38294739276

Reproduce

wiretruth serve ollama-openai/request.max-tokens-ollama-openai

Plain-library snippet from the shim manifest.

OllamaModel(model, provider=OllamaProvider(base_url=WIRETRUTH_BASE_URL_OLLAMA_OPENAI, api_key=WIRETRUTH_API_KEY)); /v1 base

Upstream

reported Issue · PR

Fixed in: not specified

Existing upstream issue, not opened by wiretruth. Ollama's OpenAI-compatible endpoint ignores max_completion_tokens (v0.40.3), and OllamaModel sends only that name (OllamaProvider keeps the OpenAI profile default, openai_chat_supports_max_completion_tokens=True), so the limit never applies. The root cause is in Ollama, as for any-llm; the check reads the request, so retire or rewrite the scenario when Ollama ships the fix.

History

Version	Grade
2.61.0	✗ WRONG
2.56.0	✗ WRONG