responses.parse() fails with GPT-5 reasoning/verbosity parameters

[oxynad]

Description

The responses.parse() method throws an error when combining Pydantic structured outputs with GPT-5's
new reasoning and text parameters.

Environment

• openai-python: 1.99.3 (AsyncOpenAI and OpenAI)
• Model: gpt-5

Minimal reproduction

  from pydantic import BaseModel
  from openai import AsyncOpenAI

  class BookSummary(BaseModel):
      title: str
      summary: str
      rating: int

  client = AsyncOpenAI()

This fails

  response = await client.responses.parse(
      model="gpt-5",
      input=[
          {"role": "user", "content": "Summarize this book..."}
      ],
      text_format=BookSummary,
      reasoning={"effort": "minimal"},
      text={"verbosity": "medium"}
  )

Error

Error code: 400 - {'error': {'message': "Unsupported parameter: 'reasoning.effort' is not supported
with this model.", 'type': 'invalid_request_error', 'param': 'reasoning.effort'}}

Expected behavior

responses.parse() should support GPT-5's reasoning/verbosity parameters alongside Pydantic models.

Impact

Cannot use structured outputs with GPT-5's performance optimization features, forcing a choice between
type safety or the new model capabilities.

To Reproduce

  client = AsyncOpenAI()
  from pydantic import BaseModel
  from openai import AsyncOpenAI

  class BookSummary(BaseModel):
      title: str
      summary: str
      rating: int
  response = await client.responses.parse(
      model="gpt-5",
      input=[
          {"role": "user", "content": "Summarize this book..."}
      ],
      text_format=BookSummary,
      reasoning={"effort": "minimal"},
      text={"verbosity": "medium"}
  )

OS

macOS Sequoia 15.5

Python version

Python v3.13.5

Library version

openai v1.99.3