MiMo-v2.5-Pro-UltraSpeed: 1T model with 1000 tokens per second

Re:MiMo-v2.5-Pro-UltraSpeed: 1T model with 1000 tokens per s... by nl on Monday June 08, 2026 @07:44PM (#48454045)

Actually in this case they possibly are estimates.

It's been known for some years[1] that LLMs do regression in-context. Frontier models have been trained against many, many issue text that include task break downs and estimates.

[1] https://arxiv.org/html/2409.04318v1

[ Parent | on HN ]

Re:MiMo-v2.5-Pro-UltraSpeed: 1T model with 1000 tokens per s... by kube-system on Monday June 08, 2026 @09:33PM (#48455025)

Interesting. So it may have learned how to estimate as a human but doesn’t understand that it doesn’t operate at that speed :D

I wonder if there’s a reasonable way to give an llm parameters that give it a concept of its own execution speed. Seems that could be useful for multiple purposes

[ Parent | on HN ]

Re:MiMo-v2.5-Pro-UltraSpeed: 1T model with 1000 tokens per s... by nl on Tuesday June 09, 2026 @02:12AM (#48457172)

Yes, it's entirely possible to do that via RL. It'd be a fun little project you could do for less than $100 on a small LLM actually.

[ Parent | on HN ]