Hey thanks for the report.
The thing is that DeepSeek V4 both deepseek-v4-pro and deepseek-v4-flash uses thinking mode, which requires returning reasoning_content back to the API during multi-turn tool calls. Cursor can’t do that yet, so the DeepSeek API returns this error:
The reasoning_content in the thinking mode must be passed back to the API.
There are already a few recent reports about this on the forum:
- The deepseek-v4-pro api cannot be used normally
- Error when i use deepseek-v4-flash
- DeepSeek V4: context limited to 200K + reasoning_content error
There is also a separate feature request to add compatibility. Can you drop a like to help raise priority: Compatibility with DeepSeek model's design to return reasoning content after tool calls?
What you can do right now:
- Community proxy. One of the users made a compatibility layer that wraps the DeepSeek API so it works with Cursor:
GitHub - yxlao/deepseek-cursor-proxy: Fixes reasoning_content issue of DeepSeek V4, supprting deepseek-v4-pro and deepseek-v4-flash · GitHub
Run it locally, then setOverride OpenAI Base URLto the proxy address instead ofapi.deepseek.com.