Xiaomi released MiMo-V2.6-Pro-RL and Flash-RL as open weights under MIT, plus its RL training environments. The company's own cost and benchmark figures are unverified.
Xiaomi's MiMo team open-sourced MiMo-V2.6-Pro-RL and MiMo-V2.6-Flash-RL on 22 September 2026, along with a smaller distilled model, MiMo-V2.6-Distill-Qwen-9B, and the reinforcement-learning infrastructure used to train them. All three are released under an MIT license, which permits commercial use, and are available on Hugging Face and through Xiaomi's own MiMo API platform.
Beyond publishing weights, Xiaomi released the underlying technical report, more than 7,000 reinforcement-learning task environments, and the end-to-end RL framework used to train the models. That level of detail lets outside researchers check whether a benchmark gain came from the model itself, the task set, the grader, or the training framework — a distinction most model releases leave opaque.
The numbers are Xiaomi's own
Xiaomi reports that MiMo-V2.6-Pro tops Artificial Analysis's open-weights intelligence index, and says the RL training runs cost roughly $2.62 million for Pro and $850,000 for Flash, completed in under six days. The company also reports large jumps on internal held-out benchmarks after RL training. These figures — the ranking, the cost estimates and the benchmark deltas — are all company-reported and have not been independently reproduced; treat them as Xiaomi's account of its own training run rather than a verified industry ranking.
What is actually confirmed
The weights, the MIT license, the technical report and the RL environments are real, published artifacts that anyone can download and inspect today. That is the durable part of the story for MSPs and engineering teams evaluating open-weight models: a verifiable release paired with unverifiable performance marketing.
We use technical cookies to keep the site running. With your consent we also use analytics and marketing cookies to improve the product. You can change your mind at any time.
01
NecessaryAlways on
Required for the site to work (login, preferences, security). Always on.
02
Help us understand how the site is used, in aggregate and anonymous form.
03
Let us measure how well our campaigns perform. Off by default.
