2.79.1: OpenAI gpt-6 and later are reasoning families

[CocoRoF]

Problem

On Odysseus prod the chat default is OpenAI gpt-6-luna. Every AI agent turn failed with:

Function tools with reasoning_effort are not supported for gpt-6-luna in /v1/chat/completions. To use function tools, use /v1/responses or set reasoning_effort to 'none'.

The reasoning-family table only named o1, o3, o4 and gpt-5. gpt-6-luna matched none of them, so its tool turns stayed on Chat Completions instead of the Responses wire.

Fix

• GPT generations 5 and later are matched by number (gpt-6-luna, gpt-6.1-sol, gpt-7, gpt-10-mini). They get the Responses wire for tool turns, max_completion_tokens, and default sampling.
• A heal teaches the client. When the vendor names the max_tokens rename or refuses tools with reasoning, the model is remembered for the process.
• The tools+reasoning 400 is retried once with reasoning_effort: "none" (the vendor's documented way out). The turn that discovers the model still answers.

Verification

• tests/llm_client and tests/unit/test_openai_responses_routing.py: 301 passed. tests/unit and tests/contract: 4610 passed.
• Live check inside the Odysseus prod api container with the real gpt-6-luna account:
  • The installed version (2.65.6) and 2.79.0 both still failed the agent tool call.
  • This wheel returned stop_reason=tool_use. The NPC reply was normal.

A 2.67.1 maintenance release with the same fix follows for Odysseus, which pins the 2.6x line.

🤖 Generated with Claude Code