Xiaomi Hands Out a Trillion-Parameter Model Under the MIT License
Xiaomi has published the weights of MiMo-V2.6-Pro, the trained numbers that define the model, under the MIT license. Artificial Analysis, an evaluation firm that obtained the released files and ran them itself, scores the model 46 on version 4.3.2 of its Intelligence Index. That is first place among the 114 models the index compares it with, a class made up of open-weights systems in the largest size bracket.
The MIT license allows commercial use, and Artificial Analysis lists the weights as available for download on Hugging Face, so anyone can run the model on their own hardware.
The same page puts the model at 1 trillion parameters in total, of which 42 billion are used on any given piece of text, and gives it a context window of 1 million tokens, about 1,500 pages. It accepts text, images, speech and video, and answers in text. The firm measured its output at 54 tokens a second, slower than the average in its class.
The class is the index's own: it ranks open-weights models only against other open-weights models of a similar size, so the first place is a rank within that group, not across the index as a whole. On the same scale, the highest-scoring model the firm has measured is Anthropic's Claude Opus 5.5, a proprietary system, at 58.
TechNode, reporting Xiaomi's Chinese-language announcement on Sept. 22, 2026, says the series also includes a faster model, MiMo-V2.6-Flash, and a smaller one, along with the company's research materials on the training method behind them.
