fix: add reasoning_content field to fix pydantic validation error

[mwien]

A minimal fix to avoid Pydantic validation errors in case reasoning content is returned in the orchestration response. Later we should add convenience functionalities around reasoning content, but this is a larger project and imo out-of-scope of this fix PR.

I quickly checked manually that this fixes the issue. We should likely add some e2e/integration tests to ensure correctness (but those have been flaky recently). In case this PR is not urgent I can take care of this before merging.

[ZhongpinWang]

Some tests + changelog would be nice. The fix is self-explanatory and looks good.

[mwien]

Added some minimal unit testing for now, previousy there was no unit testing for ChatResponseMessage unless I missed something. So maybe in a separate PR more testing can be added.

Changelog: where should I log the changes?

[ZhongpinWang]

Added some minimal unit testing for now, previousy there was no unit testing for ChatResponseMessage unless I missed something. So maybe in a separate PR more testing can be added.

Changelog: where should I log the changes?

Maybe @marikaner @alpkom can help regarding changelog?

[yamaceay]

lgtm, and also I successfully ran the following e2e test:

from gen_ai_hub.proxy.core import get_proxy_client
from gen_ai_hub.orchestration_v2.service import OrchestrationService
from gen_ai_hub.orchestration_v2.models.llm_model_details import LLMModelDetails 
from gen_ai_hub.orchestration_v2.models.message import ReasoningBlock, SystemMessage, UserMessage
from gen_ai_hub.orchestration_v2.models.template import Template, PromptTemplatingModuleConfig
from gen_ai_hub.orchestration_v2.models.config import OrchestrationConfig, ModuleConfig

client = get_proxy_client(proxy_version="gen-ai-hub")
running = [d for d in client.ai_core_client.deployment.query(scenario_id="orchestration").resources if d.status.value == "RUNNING"]
api_url = running[0].deployment_url
service = OrchestrationService(api_url)

llm = LLMModelDetails(
    name="anthropic--claude-4.5-sonnet",
    params={
        "temperature": 1.0,
        "thinking": {"type": "enabled", "budget_tokens": 1024},
    }
)
template = Template(template=[
    SystemMessage(content="You are a helpful assistant."),
    UserMessage(content="What is 17 * 23?"),
])
ptm = PromptTemplatingModuleConfig(prompt=template, model=llm)
config = OrchestrationConfig(modules=ModuleConfig(prompt_templating=ptm))
response = service.run(config=config)

parsed_msg = response.final_result.choices[0].message
if parsed_msg.reasoning_content:
    print("Pydantic model correctly populated reasoning_content")
    assert isinstance(parsed_msg.reasoning_content[0], ReasoningBlock)
    print("isinstance(block, ReasoningBlock) = True")
else:
    print("Pydantic model did NOT populate reasoning_content despite raw data present")

[mwien]

Thanks! Should we add this to the e2e tests?

[yamaceay]

Thanks! Should we add this to the e2e tests?

In my opinion, unit tests also cover the minimal behavior change, so that's fine I guess. Adding it to e2e might increase the integration test duration unnecessarily, but what do you think?

[mwien]

I was just wondering why this issue wasn't detected in e2e testing earlier. But maybe let's only add it once we fully support reasoning/thinking.