fix(ai): use reasoning_effort for Mistral Small 4 closes #3338

This commit is contained in:
Mario Zechner
2026-04-17 22:44:03 +02:00
parent e6f473f432
commit 62778a82d5
6 changed files with 108 additions and 17 deletions

View File

@@ -5,6 +5,7 @@
### Fixed
- Fixed Anthropic and Bedrock adaptive-thinking payload tests to expect the default `display: "summarized"` field when reasoning is enabled.
- Fixed Mistral Small 4 reasoning requests to use `reasoning_effort` instead of `prompt_mode`, restoring default thinking support for `mistral-small-2603` and `mistral-small-latest` ([#3338](https://github.com/badlogic/pi-mono/issues/3338))
- Fixed `qwen-chat-template` OpenAI-compatible requests to set `chat_template_kwargs.preserve_thinking: true`, preserving prior Qwen thinking across turns so multi-turn tool calls keep their arguments instead of degrading to empty `{}` payloads ([#3325](https://github.com/badlogic/pi-mono/issues/3325))
## [0.67.6] - 2026-04-16