Hey, this is a known BYOK bug for providers that require reasoning_content passback like Kimi K2.6, DeepSeek V4, and similar. The first turn works, but on the next tool calls the provider returns 400 “thinking is enabled but reasoning_content is missing”. The same issue is discussed here: Kimi 2.6 custom model - Provider returned error: {"error":{"message":"thinking is enabled but reasoning_content is missing
We’re tracking it, but there’s no ETA for a fix yet. I grabbed the Request ID and I’ll link it to the ticket.
As a workaround, you can temporarily use the non-thinking Kimi option where reasoning passback isn’t required, or use another model in agent mode with tool calls. I’ll follow up once there’s an update.