What happened
- On September 22, 2026, Xiaomi released the MiMo-V2.6 series with open weights, in three variants: Pro, Flash and Pro-UltraSpeed.
- The models are natively omnimodal (text, image, video and audio) and work with context windows of one million tokens. The Pro-UltraSpeed variant generates up to 20 times faster than Pro.
- On the Artificial Analysis index, MiMo-V2.6-Pro scored 46.32 and was the top-ranked open model at the time of launch. On AutomationBench it scored 53.1 versus 50.3 for Claude Opus 5, and on Terminal Bench 2.1 it reached 89.9 versus 89.1.
- On OpenRouter, Flash costs $0.14 per million input tokens and $0.28 for output; Pro, $0.435 and $0.87.
Why it matters
- An open-weights model that performs like a closed one changes where data can live. Anyone working with sensitive information in Chile or the region can host it on their own infrastructure and stop exporting those records to process them.
- Flash’s price puts the entry cost an order of magnitude below the frontier models announced the same day. For volume tasks (classifying, transcribing, labeling), the comparison is no longer about quality but about integration.
- The three benchmark wins are narrow and come from the manufacturer itself. A 0.8-point difference on Terminal Bench doesn’t support an architecture decision; it does justify running your own test with your own data.
The number
46.32 points on Artificial Analysis’s intelligence index, the highest score among open models at the time of launch. It’s the number Xiaomi uses to enter the conversation.
Context
The open ecosystem has been changing owners and prices: Nvidia bought Hugging Face for $12.93 billion, and Chinese providers are pushing rates down, as DeepSeek’s time-of-day pricing showed. Xiaomi comes in through the door of open weights and a long context window.
What’s next
- The models are already available in MiMo Studio, Desktop, Code and on OpenRouter.
- No timelines announced for later versions or for commercial support outside China.
Bottom line
On the same day two U.S. labs cut prices in half, a phone maker published the weights. They’re two different ways of competing for the same budget.
Sources
Edited by Rodrigo Cornejo. How we select and verify each fact is in who writes.
