Reasoning summaries should be send in the language of the user

API
1

Hi everyone,

I’m currently building an agent workflow that relies on streamed reasoning summaries emitted through stream_agent_response (from the Responses API / Agents SDK). It works great—except for one thing:

All reasoning summaries are currently streamed in English, even when the user’s language is clearly different (e.g. French).

This creates an odd UX where:

  • The assistant response is in French,

  • But the thought process / reasoning summaries are in English,

  • Which breaks immersion and confuses non-English-speaking users.

Why It Matters

  • Multilingual users deserve coherent experiences, especially when reasoning is displayed to them in real time.

  • For accessibility and transparency, reasoning summaries should adopt the same language as the user’s input or system instructions.

  • This also aligns with how OpenAI models generally mirror user language in normal chat completions.

Proposed Behavior

  • If the user prompt or conversation context indicates a language (e.g. "lang":"fr" or simply detected French text),
    → the model should stream reasoning summaries in that language as well.

Even a simple heuristic like “use the language of the last user message” would cover 95% of cases.

Question for the OpenAI Team

  • Is there already a way to enforce reasoning summary language via system prompt or model setting?

  • If not, could this be added as either:

    • A parameter (e.g. reasoning_language="auto" | "fr" | "en"), or

    • Automatic detection based on the conversation language?

1 Like