Been playing around with minimax 2.5 with OH and am getting great results. But seems that this new "interleaved thinking" concept is changing how the output works. In openwebui, the thinking remains within the "thinking block" in the interface. But it seems that, unlike other reasoning models in openhands, the thinking portion for minimax is being written directly to the output instead of the "🤔 thinking" block. Is this potentially problematic for the actual agent? Or just a ui inconvenience for now?

Probably UI, but I'm not totally sure. Maybe if you can reproduce a problem or weird thing or prompt/response logs from the LLM with the SDK, please file an issue