Encrypted content is not supported for GPT-5 though Azure

[alibeyram]

Initial Checks

• I confirm that I'm using the latest version of Pydantic AI
• I confirm that I searched for my issue in https://github.com/pydantic/pydantic-ai/issues before opening this issue

Description

when running gpt-5 through Azure you get the following error

ModelHTTPError: status_code: 400, model_name: gpt-5, body: {'message': 'Encrypted content is not supported with this model.', 'type': 'invalid_request_error', 'param': 'include', 'code': None}

Example Code

import os

from openai import AsyncAzureOpenAI
from pydantic import BaseModel
from pydantic_ai import Agent, NativeOutput
from pydantic_ai.models.openai import OpenAIResponsesModel
from pydantic_ai.providers.openai import OpenAIProvider


class CalendarEvent(BaseModel):
    name: str
    date: str
    participants: list[str]


client = AsyncAzureOpenAI(
    azure_endpoint=os.environ["AZURE_OPENAI_ENDPOINT"],
    api_version=os.environ["OPENAI_API_VERSION"],
    api_key=os.getenv("AZURE_OPENAI_API_KEY"),
)

model = OpenAIResponsesModel(
    model_name="gpt-5",
    provider=OpenAIProvider(openai_client=client),
)

agent = Agent(
    model=model,
    tools=[],
    system_prompt="Extract the event information.",
    output_type=NativeOutput(CalendarEvent),
)
result = await agent.run(
    [
        "Alice and Bob are going to a science fair on Friday.",
    ]
)

Python, Pydantic AI & LLM client version

python -V
Python 3.13.7

import pydantic_ai
pydantic_ai.__version__
'1.0.2'

[alibeyram]

@DouweM suggested the following workaround that works

import os
from dataclasses import replace

from openai import AsyncAzureOpenAI
from pydantic import BaseModel
from pydantic_ai import Agent, NativeOutput
from pydantic_ai.models.openai import OpenAIModelProfile, OpenAIResponsesModel
from pydantic_ai.profiles.openai import openai_model_profile
from pydantic_ai.providers.openai import OpenAIProvider


class CalendarEvent(BaseModel):
    name: str
    date: str
    participants: list[str]


client = AsyncAzureOpenAI(
    azure_endpoint=os.environ["AZURE_OPENAI_ENDPOINT"],
    api_version=os.environ["OPENAI_API_VERSION"],
    api_key=os.getenv("AZURE_OPENAI_API_KEY"),
)


profile = replace(
    OpenAIModelProfile.from_profile(openai_model_profile("gpt-5")),
    openai_supports_encrypted_reasoning_content=False,
)

model = OpenAIResponsesModel(
    model_name="gpt-5",
    provider=OpenAIProvider(openai_client=client),
    profile=profile,
)

agent = Agent(
    model=model,
    tools=[],
    system_prompt="Extract the event information.",
    output_type=NativeOutput(CalendarEvent),
)
result = await agent.run(
    [
        "Alice and Bob are going to a science fair on Friday.",
    ]
)

[DouweM]

@alibeyram Thanks for the report. I can disable encrypted_content for all Azure OpenAI, but interestingly https://learn.microsoft.com/en-us/azure/ai-foundry/openai/reference-preview-latest#request-body-14 does document the include=[reasoning.encrypted_content] option as being available on their "REST API v1 preview". Per https://learn.microsoft.com/en-us/azure/ai-foundry/openai/api-version-lifecycle?tabs=key#api-evolution, you're currently using the "last generation API" (using the AzureOpenAI client), and they now have a new "next generation API" that works with the regular OpenAI client and base_url https://YOUR-RESOURCE-NAME.openai.azure.com/openai/v1/.

Would you mind trying if everything works out of the box with that new API? Then I can disable encrypted_content only for the last gen API, and not for Azure OpenAI as a whole.

[cmorrison3r3]

I had this encrypted error. Re-run your build and try Pydanticai 1.0.7. Works for me now on Render with gpt4.1

[alibeyram]

@DouweM

gpt-4.1 works without any issue. this is specifically gpt-5 problem

this is what I ran

import os
from pydantic import BaseModel
from pydantic_ai import Agent, NativeOutput
from pydantic_ai.models.openai import OpenAIResponsesModel
from pydantic_ai.providers.openai import OpenAIProvider


class CalendarEvent(BaseModel):
    name: str
    date: str
    participants: list[str]


model = OpenAIResponsesModel(
    model_name="gpt-5",
    provider=OpenAIProvider(
        base_url="https://OUR-RESOURCE-NAME.cognitiveservices.azure.com/openai/v1",
        api_key=os.getenv("AZURE_OPENAI_API_KEY"),
    ),
)

agent = Agent(
    model=model,
    tools=[],
    system_prompt="Extract the event information.",
    output_type=NativeOutput(CalendarEvent),
)
result = await agent.run(
    [
        "Alice and Bob are going to a science fair on Friday.",
    ]
)

same error

ModelHTTPError: status_code: 400, model_name: gpt-5, body: {'message': 'Encrypted content is not supported with this model.', 'type': 'invalid_request_error', 'param': 'include', 'code': None}

again if I change the model_name to gpt-4.1 it works. Given that I would say it is a bug on their side and probably they are gonna fix it