Xiaomi released MiMo-V2.6 Pro as an open-weight omnimodal system under an MIT license capable of processing text, image, video, and audio. The flagship model scored 46.32 on the Artificial Analysis Intelligence Index, the highest result for any open model to date. Its Mixture-of-Experts architecture comprises 1.02 trillion total parameters and 42 billion active parameters, with pricing set at $0.435 per 1 million input tokens and $0.87 per 1 million output tokens.
The current Pro version increased its intelligence score from 26 in the previous V2.5 iteration to 46 after 30 reinforcement learning steps. This final training run lasted 5 days on a 10,000 GPU cluster and was shipped less than one week after completion. Along with the model weights, Xiaomi is providing a corresponding Flash variant, an end-to-end RL framework, and approximately 7,000 RL training environments.
Key sources
- SOURCE@victormustar“Pro scores 46.32 on the Artificial Analysis Intelligence Index, the highest of any open model to date”x.com
- SUPPORT@artificialanlys“MoE model with 1.02T total parameters and 42B active parameters”x.com
- SUPPORT@teortaxestex“This RL run had cost peanuts, 5 days of a ≈10K GPU cluster”x.com
- SUPPORT@teortaxestex“Pro performs on par with Claude Opus 5 and GPT-5.6 Sol across most agent benchmarks”x.com
- SUPPORT@eliebakouch“release ~7k RL training data and the framework leading to this top 6 model on AA”x.com
- SUPPORT@nrehiew_“pacing the frontier using only 30 steps”x.com
- SOURCEmarketbrief.now
- SOURCE@arena“scores 1628 pts (AutoEval), tying Claude Fable 5 (High) and just ahead of Hy4-preview (1624 pts)”x.com