Hi Team,
I’m a developer from Z.ai, and we’ve been integrating our GLM models with Cursor via the Chat Completions endpoint. We’ve run into two integration issues and would really appreciate your support.
- Context Size Misconfiguration (1M vs 200K)
Currently, when accessing GLM-5 and GLM-4.7 through the Chat Completions endpoint, Cursor displays a 1M context window.
However, the actual supported context length of the models is around 200K tokens.
Because of this mismatch, conversations overflow after several turns, resulting in context length errors. It would be extremely helpful if Cursor could configure the correct maximum context size for the GLM models to avoid this issue.
- Thinking / Reasoning Mode Adaptation
There is also a compatibility issue regarding reasoning mode.
In the GLM API design, reasoning output is returned in:
choices.message.reasoning_content
It is not embedded inside choices.message.content using tags.
Currently, when users connect third-party GLM APIs to Cursor, the reasoning capability cannot function properly because Cursor expects reasoning content inside choices.message.content.
We would greatly appreciate it if Cursor could support parsing choices.message.reasoning_content for GLM models so that the reasoning mode works correctly.
⸻
We’re very much looking forward to deeper collaboration between z.ai and Cursor. Thank you in advance for your help and support.
Best regards,
Chao Gong