✓ PASS

pydantic-ai 2.61.0

usage.reasoning-tokens · anthropic

Requirements

Id	Level	Status	Evidence
total-reconciles	MUST	SATISFIED	usage.total=400 (primary); truth.usage.total=400
reasoning-reported	SHOULD	SATISFIED	usage.reasoning=312 (primary); truth.usage.reasoning=312

What the provider sent

Finish: stop
Raw finish: end_turn
Text: 11 pots contain three seedlings: 11 times 3 plus 7 times 2 equals 47.
Tool calls: []
Usage: { "alternatives": {}, "cached_input": null, "input": 52, "output": 348, "raw": { "input_tokens": 52, "output_tokens": 348, "output_tokens_details": { "thinking_tokens": 312 } }, "reasoning": 312, "total": 400 }
Stream completed: null
Error: null

Wire excerpt

status: 200
content-type: application/json
body (json):
{
  "id": "msg_reasoning_example",
  "type": "message",
  "role": "assistant",
  "content": [
    {
      "type": "thinking",
      "thinking": "",
      "signature": "documented-placeholder"
    },
    {
      "type": "text",
      "text": "11 pots contain three seedlings: 11 times 3 plus 7 times 2 equals 47."
    }
  ],
  "model": "claude-sonnet-5-5",
  "stop_reason": "end_turn",
  "stop_sequence": null,
  "stop_details": null,
  "usage": {
    "input_tokens": 52,
    "output_tokens": 348,
    "output_tokens_details": {
      "thinking_tokens": 312
    }
  }
}

What the library reported

Outcome

returned

Finish

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

Text

Value: 11 pots contain three seedlings: 11 times 3 plus 7 times 2 equals 47.
Surface: primary

Tool calls

[]

Usage

{
  "cached_input": 0,
  "input": 52,
  "output": 348,
  "reasoning": 312,
  "surface": {
    "cached_input": "primary",
    "input": "primary",
    "output": "primary",
    "reasoning": "primary",
    "total": "primary"
  },
  "total": 400
}

Stream

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

Error

null

Unsupported reason

null

Full observation JSON

{
  "elapsed_ms": 255.76307252049446,
  "error": null,
  "finish": {
    "raw": "end_turn",
    "source": "ModelResponse.finish_reason",
    "surface": "primary",
    "value": "stop"
  },
  "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": "11 pots contain three seedlings: 11 times 3 plus 7 times 2 equals 47."
  },
  "tool_calls": [],
  "unsupported_reason": null,
  "usage": {
    "cached_input": 0,
    "input": 52,
    "output": 348,
    "reasoning": 312,
    "surface": {
      "cached_input": "primary",
      "input": "primary",
      "output": "primary",
      "reasoning": "primary",
      "total": "primary"
    },
    "total": 400
  }
}

Fixture

documented anthropic/usage.reasoning-tokens

Provenance	Value
Source	https://platform.claude.com/docs/en/build-with-claude/thinking-steering-and-cost#pricing
Notes	Documented shapes, not a recording. The pricing example supplies output_tokens 348 and output_tokens_details.thinking_tokens 312; output_tokens includes thinking. Message envelope and end_turn follow the Messages API reference (https://platform.claude.com/docs/en/api/messages/create). Model, id, input_tokens and greenhouse answer are filled in; the combined response is assumed, not documented verbatim. The empty thinking block follows the omitted-display shape (https://platform.claude.com/docs/en/build-with-claude/thinking#control-how-thinking-is-shown); its signature is a placeholder. Request built with providers.anthropic.build_request: adaptive thinking, output_config.effort medium and max_tokens 800. Adaptive thinking has no manual budget_tokens minimum; thinking and text share the output cap. No request was sent.

Provenance

Value

Source

https://platform.claude.com/docs/en/build-with-claude/thinking-steering-and-cost#pricing

Notes

Documented shapes, not a recording. The pricing example supplies output_tokens 348 and output_tokens_details.thinking_tokens 312; output_tokens includes thinking. Message envelope and end_turn follow the Messages API reference (https://platform.claude.com/docs/en/api/messages/create). Model, id, input_tokens and greenhouse answer are filled in; the combined response is assumed, not documented verbatim. The empty thinking block follows the omitted-display shape (https://platform.claude.com/docs/en/build-with-claude/thinking#control-how-thinking-is-shown); its signature is a placeholder. Request built with providers.anthropic.build_request: adaptive thinking, output_config.effort medium and max_tokens 800. Adaptive thinking has no manual budget_tokens minimum; thinking and text share the output cap. No request was sent.

Reproduce

wiretruth serve anthropic/usage.reasoning-tokens

Plain-library snippet from the shim manifest.

AnthropicModel(model, provider=AnthropicProvider(base_url=WIRETRUTH_BASE_URL_ANTHROPIC, api_key=WIRETRUTH_API_KEY)); host root

Upstream

No matching upstream report.

History

Version	Grade
2.61.0	✓ PASS
2.56.0	not covered