Chinese reasoning models may need hidden memory on turn two
DeepSeek, Kimi, GLM, MiMo, and MiniMax can answer the first question correctly, then fail with HTTP 400 on the follow-up. The problem may be the chat client, not the model. These send a hidden field called reasoning_content, which works like internal notes made before the visible answer.
The model may need that hidden field returned on the next turn so it can validate the . OpenCode, Cursor, Claude Code, and VS Code Copilot can drop reasoning_content because it is not part of the . A one-turn test can pass, while real users fail on turn two.
The issue has reportedly been visible since January, and the related plugin has patched 12,551 messages.
Key points
- DeepSeek, Kimi, GLM, MiMo, and MiniMax may fail on the second turn with HTTP 400.
- The likely issue is a client dropping reasoning_content, not a broken model.
- tools may ignore the field because it is not part of that format.
- A good test should include at least one follow-up question, not only the first reply.
- If turn two fails, check whether hidden reasoning fields are preserved in the .