Sources
Xiaomi's MiMo-V2.6-Pro takes the open-weights crown at 1T total / 42B active, MIT licensed, trained for $2.62M
Xiaomi released and open-sourced the MiMo-V2.6 series on September 21-22, and Artificial Analysis scored MiMo-V2.6-Pro at 46 on its Intelligence Index, ranking it #1 of 114 and ahead of Kimi K3 (44) and GLM-5.3 (45), with leading closed models at 53. The flagship is a natively omnimodal MoE with 1.02T total and 42B active parameters, a 1M context window, MIT-licensed weights on Hugging Face, and pricing of $0.435 per 1M input and $0.87 per 1M output. For builders this is the first frontier-class open-weights model from a phone maker rather than one of the six established Chinese AI labs, and the accompanying MiMo-V2.6-Distill-Qwen-9B gives a small local option from the same recipe.
↳ Follow the thread