88Z AI, 22 September — Xiaomi has released and open-sourced the Xiaomi MiMo-V2.6 series, featuring two native full-modal models: Pro and Flash. After extensive reinforcement learning training, the Pro version surpasses several well-known models in comprehensive intelligence index, becoming the strongest open-source model, though it still lags behind top closed-source counterparts.
At a glance
- MiMo-V2.6 Pro becomes the strongest open-source full-modal model after RL training
- Outperforms Kimi K3 and others but still trails top closed-source models
The update
Xiaomi has officially launched and open-sourced the Xiaomi MiMo-V2.6 series, which includes two native full-modal versions: Pro and Flash. These models underwent approximately six days of large-scale live reinforcement learning training, optimized using 750,000 trajectory data points. Tests indicate that the Pro model achieved a score of 46 on the AA Comprehensive Intelligence Index, outperforming Kimi K3 and Qwen3.8 Max, thereby establishing its status as the current strongest open-source model. Despite this strong performance, a technical gap remains when compared to top closed-source models such as Claude Fable 5.1 and GPT-6 Astra.
Sources & editorial note
This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
- Xiaomi MiMoSource published
Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.