[diegosouzapw]

Summary

• OpenAI's Chat Completions endpoint rejects GPT-5.6 requests that combine function tools with an active reasoning_effort — 400 "Function tools with reasoning_effort are not supported for <model> in /v1/chat/completions. Please use /v1/responses instead." Any agentic client hitting openai/gpt-5.6-sol with tools + reasoning got the 400 and burned a combo fallback attempt.
• Root cause: the openai (API-key) registry entries for gpt-5.6 / -sol / -terra / -luna were missing the per-model targetFormat tag, so every request went to /v1/chat/completions.
• Fix reuses the mechanism OmniRoute already has: the per-model targetFormat: "openai-responses" override that already routes gpt-5.5-pro / gpt-5.4-pro (fix(providers): Various OpenAI model returns [404] error due to endpoint #5842). It drives BOTH the outbound URL (DefaultExecutor.buildUrl → api.openai.com/v1/responses) and the body translation (resolveChatCoreTargetFormat → openai-responses). No new transport table or routing branch needed.

Attribution

Thanks to @Jordannst for the original implementation.

Changes

• open-sse/config/providers/shared.ts — add targetFormat: "openai-responses" to GPT_5_6_API_CAPABILITIES (consumed only by the 4 public OpenAI GPT-5.6 catalog entries). Scoped to the public OpenAI API catalog; the codex provider has its own Responses transport and is untouched.
• tests/unit/openai-gpt56-responses-routing.test.ts — new regression test.

Test plan

• targeted unit test (new, fails before / passes after)

  Before (on release/v3.8.49):

  ✖ getModelTargetFormat routes the public OpenAI GPT-5.6 family through Responses
    AssertionError: gpt-5.6 must target openai-responses
    + actual: null   - expected: 'openai-responses'
  ✖ DefaultExecutor builds the /v1/responses URL for gpt-5.6-sol
    + actual: 'https://api.openai.com/v1/chat/completions'
    - expected: 'https://api.openai.com/v1/responses'
  

  After: ℹ tests 4 / ℹ pass 4 / ℹ fail 0

• npm run typecheck:core — clean

• npx eslint on changed files — clean (exit 0)

Negative coverage included: gpt-5.4 still resolves to null / /v1/chat/completions.

---

⚠️ Reviewer note — competing approach in flight

Another in-flight change targets the same root cause (GPT-5.6 + tools + reasoning) with the opposite strategy, and the two should not land together without a decision:

• This PR preserves the client's reasoning_effort by routing the request to the Responses API.
• The other approach (tests/unit/gpt5-tools-reasoning-guard.test.ts, currently uncommitted WIP) drops the client's reasoning when tools are present.

They are mutually exclusive in intent: one keeps reasoning by changing the endpoint, the other silently discards it. Worth picking one deliberately rather than merging both.

FWIW that WIP test currently does not import — it pulls stripGpt5ReasoningWhenTools from open-sse/services/gpt5SamplingGuard.ts, which does not export that symbol.

[gemini-code-assist]

Warning

You have reached your daily quota limit. Please wait up to 24 hours and I will start processing your requests again!

[diegosouzapw]

Aprovado tecnicamente — fail-without-fix confirmado (2/4 falham sem o fix) e sem colisão de arquivo com #7012/#7209 (constantes/arquivos diferentes). Mas o PR já sinaliza corretamente o ponto crítico: o #7101 ataca a MESMA causa-raiz com estratégia OPOSTA (descarta reasoning_effort em vez de re-rotear) e não exclui gpt-5.6 do seu escopo — se os dois forem mergeados sem decisão, o comportamento final de gpt-5.6+tools+reasoning depende da ordem de execução em chatCore.ts e pode anular este fix. Recomendo decidir ANTES de mergear ambos: ou #7101 exclui explicitamente os modelos com targetFormat já resolvido para 'openai-responses', ou só um dos dois entra.