response_format: response_format.type `json_schema`

[summerHearts]

Please read this first

• Have you read the docs?Agents SDK docs YES
• Have you searched for related issues? Others may have had similar requesrs YES

Question

import asyncio
import random
from typing import Any
from pydantic import BaseModel
import asyncio
from dotenv import load_dotenv
from agents import Agent, Runner, AsyncOpenAI, OpenAIChatCompletionsModel, ModelSettings
from agents import set_default_openai_client, set_tracing_disabled
from agents import Agent, RunContextWrapper, RunHooks, Runner, Tool, Usage, function_tool
import os

load_dotenv()

deepseek_api_key = os.getenv('DEEPSEEK_API_KEY')

custom_client = AsyncOpenAI(
    api_key=os.getenv('DEEPSEEK_API_KEY'),
    base_url="https://api.deepseek.com"
)
set_tracing_disabled(True)

set_default_openai_client(custom_client)


class ExampleHooks(RunHooks):
    def __init__(self):
        self.event_counter = 0

    def _usage_to_str(self, usage: Usage) -> str:
        return f"{usage.requests} requests, {usage.input_tokens} input tokens, {usage.output_tokens} output tokens, {usage.total_tokens} total tokens"

    async def on_agent_start(self, context: RunContextWrapper, agent: Agent) -> None:
        self.event_counter += 1
        print(
            f"### {self.event_counter}: Agent {agent.name} started. Usage: {self._usage_to_str(context.usage)}"
        )

    async def on_agent_end(self, context: RunContextWrapper, agent: Agent, output: Any) -> None:
        self.event_counter += 1
        print(
            f"### {self.event_counter}: Agent {agent.name} ended with output {output}. Usage: {self._usage_to_str(context.usage)}"
        )

    async def on_tool_start(self, context: RunContextWrapper, agent: Agent, tool: Tool) -> None:
        self.event_counter += 1
        print(
            f"### {self.event_counter}: Tool {tool.name} started. Usage: {self._usage_to_str(context.usage)}"
        )

    async def on_tool_end(
        self, context: RunContextWrapper, agent: Agent, tool: Tool, result: str
    ) -> None:
        self.event_counter += 1
        print(
            f"### {self.event_counter}: Tool {tool.name} ended with result {result}. Usage: {self._usage_to_str(context.usage)}"
        )

    async def on_handoff(
        self, context: RunContextWrapper, from_agent: Agent, to_agent: Agent
    ) -> None:
        self.event_counter += 1
        print(
            f"### {self.event_counter}: Handoff from {from_agent.name} to {to_agent.name}. Usage: {self._usage_to_str(context.usage)}"
        )


hooks = ExampleHooks()

###

@function_tool
def random_number(max: int) -> int:
    """Generate a random number up to the provided max."""
    return random.randint(0, max)


@function_tool
def multiply_by_two(x: int) -> int:
    """Return x times two."""
    return x * 2


class FinalResult(BaseModel):
    number: int


multiply_agent = Agent(
    name="Multiply Agent",
    instructions="Multiply the number by 2 and then return the final result.",
    model=OpenAIChatCompletionsModel(
        model="deepseek-chat",
        openai_client=custom_client,
    ),
    model_settings=ModelSettings(temperature=0.7),
    tools=[multiply_by_two],
    output_type=FinalResult,
)


start_agent = Agent(
    name="Start Agent",
    instructions="Generate a random number. If it's even, stop. If it's odd, hand off to the multipler agent.",
    model=OpenAIChatCompletionsModel(
        model="deepseek-chat",
        openai_client=custom_client,
    ),
    model_settings=ModelSettings(temperature=0.7),
    tools=[random_number],
    output_type=FinalResult,
    handoffs=[multiply_agent],
)


async def main() -> None:
    user_input = input("Enter a max number: ")
    await Runner.run(
        start_agent,
        hooks=hooks,
        input=f"Generate a random number between 0 and {user_input}.",
    )

    print("Done!")


if __name__ == "__main__":
    asyncio.run(main())

error messages

Enter a max number: 30
### 1: Agent Start Agent started. Usage: 0 requests, 0 input tokens, 0 output tokens, 0 total tokens
Traceback (most recent call last):
  File "/Users/mars.yao/workSpace/openai-agents-python/examples/basic/lifecycle_example.py", line 129, in <module>
    asyncio.run(main())
  File "/opt/anaconda3/envs/openai-agents/lib/python3.11/asyncio/runners.py", line 190, in run
    return runner.run(main)
           ^^^^^^^^^^^^^^^^
  File "/opt/anaconda3/envs/openai-agents/lib/python3.11/asyncio/runners.py", line 118, in run
    return self._loop.run_until_complete(task)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/anaconda3/envs/openai-agents/lib/python3.11/asyncio/base_events.py", line 654, in run_until_complete
    return future.result()
           ^^^^^^^^^^^^^^^
  File "/Users/mars.yao/workSpace/openai-agents-python/examples/basic/lifecycle_example.py", line 119, in main
    await Runner.run(
  File "/opt/anaconda3/envs/openai-agents/lib/python3.11/site-packages/agents/run.py", line 210, in run
    input_guardrail_results, turn_result = await asyncio.gather(
                                           ^^^^^^^^^^^^^^^^^^^^^
  File "/opt/anaconda3/envs/openai-agents/lib/python3.11/site-packages/agents/run.py", line 719, in _run_single_turn
    new_response = await cls._get_new_response(
                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/anaconda3/envs/openai-agents/lib/python3.11/site-packages/agents/run.py", line 862, in _get_new_response
    new_response = await model.get_response(
                   ^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/anaconda3/envs/openai-agents/lib/python3.11/site-packages/agents/models/openai_chatcompletions.py", line 116, in get_response
    response = await self._fetch_response(
               ^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/anaconda3/envs/openai-agents/lib/python3.11/site-packages/agents/models/openai_chatcompletions.py", line 498, in _fetch_response
    ret = await self._get_client().chat.completions.create(
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/anaconda3/envs/openai-agents/lib/python3.11/site-packages/openai/resources/chat/completions/completions.py", line 2000, in create
    return await self._post(
           ^^^^^^^^^^^^^^^^^
  File "/opt/anaconda3/envs/openai-agents/lib/python3.11/site-packages/openai/_base_client.py", line 1767, in post
    return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/anaconda3/envs/openai-agents/lib/python3.11/site-packages/openai/_base_client.py", line 1461, in request
    return await self._request(
           ^^^^^^^^^^^^^^^^^^^^
  File "/opt/anaconda3/envs/openai-agents/lib/python3.11/site-packages/openai/_base_client.py", line 1562, in _request
    raise self._make_status_error_from_response(err.response) from None
openai.UnprocessableEntityError: Failed to deserialize the JSON body into the target type: response_format: response_format.type `json_schema` is unavailable now at line 1 column 483

how can i fix it

[rm-openai]

This seems to be an issue with base_url="https://api.deepseek.com". When I use custom_client without pointing to deep seek, this works fine. I'd suggest following up with the deepseek team for structured output support.

[jaijuneja]

Just to add to this for clarity, Deepseek currently supports JSON mode (where response_format is set to json_object), but not Structured Outputs (a newer evolution of JSON mode in which response_format is set to json_schema). Hopefully Deepseek will add support for Structured Outputs soon.

See the relevant page from Deepseek API docs here.

OpenAI API docs on structured outputs here.

The particular code in the Agents SDK that is causing you the error is below:

    @classmethod
    def convert_response_format(
        cls, final_output_schema: AgentOutputSchema | None
    ) -> ResponseFormat | NotGiven:
        if not final_output_schema or final_output_schema.is_plain_text():
            return NOT_GIVEN

        return {
            "type": "json_schema",
            "json_schema": {
                "name": "final_output",
                "strict": final_output_schema.strict_json_schema,
                "schema": final_output_schema.json_schema(),
            },
        }

You'll find this in the file /agents/models/openai_chatcompletions.py

The implication here is that if you're setting output_type in your agent to a Pydantic model then it won't be compatible with older OpenAI models or 3rd party models that don't support structured outputs.

[jaijuneja]

Just to add - I had a play around with this and was able to get it working with Deepseek's models but only via OpenRouter.

However, you need to be very specific in your prompt with Deepseek that it needs to output raw JSON (without any other text or code block wrapping) and provide an example of the schema, otherwise you'll get occasional validation errors when parsing the output. Deepseek also explain this in their docs.

Example code below in case helpful:

from openai import AsyncOpenAI
from agents import OpenAIChatCompletionsModel, Agent
from pydantic import BaseModel

client = AsyncOpenAI(
    api_key=OPENROUTER_API_KEY,
    base_url="https://openrouter.ai/api/v1",
)

model = OpenAIChatCompletionsModel(
    model="deepseek/deepseek-r1",
    openai_client=client
)

class TestOutput(BaseModel):
    response: str

agent = Agent(
    name="Test Agent",
    instructions="You are a helpful assistant. Your response should be in the form of JSON with the following schema: {'response': 'your response here'}. Do not include any other text and do not wrap the response in a code block - just provide the raw json output.",
    model=model,
    output_type=TestOutput
)

You can see all of OpenRouter's models that support structured outputs here.

[Avento]

https://github.com/Avento/json_schema_Forwarding/tree/main
will fix this