Just as a quick one, I have been super impressed by MiniMax-M2.5. It is probably not the best for coding, but for OpenClaw coordination, it is working super well. We have it running on an Apple Mac Studio M3 Ultra 512GB, and even the 8-bit model is running at 29-32TPS. That is in the realm of the performance we were getting with OpenRouter using Opus 4.6, depending on where it goes. That is just running it in LM Studio and 196K context. One of the fun things was that we tried using OpenClaw with OpenRouter to add the LM Studio-hosted MiniMax-M2.5, and it worked.
We also have it available in Open WebUI, and it is a great tool.
It is a fairly big jump over gpt-oss-120b, which is something very useful for agentic tasks.
We have an 8x GB10 cluster that will hopefully be set up next week and will get Kimi K2.5/ GLM 5 running on a CRS804 DDQ. I am pretty excited for that project as well.
We also have it available in Open WebUI, and it is a great tool.
It is a fairly big jump over gpt-oss-120b, which is something very useful for agentic tasks.
We have an 8x GB10 cluster that will hopefully be set up next week and will get Kimi K2.5/ GLM 5 running on a CRS804 DDQ. I am pretty excited for that project as well.
