Mistrali-large-4 thinking speed is unstable

the model thinking is fast at first, when the thinking starts, then becomes slower more and more while thinking

thinking throughput is not steady, and sometimes it stops for seconds

Maybe that’s because everyone is testing it after they deprecated K3 and GLM-5.3. I guess not everyone likes flash models.

The entire burden fell on this model, as it was the only one remaining among the mid-range and top-tier models.

What’s your opinion about this new Mistral if you used it?

rather have kimi 3 or glm 5.3 tbh.

I agree, I honestly didn’t like L4. K3.1’s weights will be released by Nov 4th, but you know NIM will add it at a later date, if they ever do. Curiously they added Large 4 really quickly.