Xiaomi Releases MiMo-V2.6-Pro and Flash as MIT-Licensed Open-Weight Models, Topping Open-Weight Ranking on Artificial Analysis Index

Overview

Xiaomi released its MiMo-V2.6 model series on September 22, 2026, according to the company’s MiMo documentation, which lists three models: mimo-v2.6-pro, mimo-v2.6-flash and mimo-v2.6-pro-ultraspeed. VentureBeat reports that MiMo-V2.6-Pro “has arrived as the top-performing open-weight model in the world on third-party benchmarking firm Artificial Analysis’ Intelligence Index, scoring 46”.

What We Know

• Architecture. VentureBeat describes Pro as having 1.02 trillion total and 42 billion active parameters, while Flash is a sparse mixture-of-experts model with 310 billion total parameters and 15 billion active during inference, per the same VentureBeat report.
• Modalities and context. SiliconANGLE reports the family accepts “text, image, video and audio in the same family, with 1 million-token context windows”.
• Benchmarks. SiliconANGLE reports that V2.6-Pro scored 46.32 on the Artificial Analysis Intelligence Index. It also reports scores of 53.1 on AutomationBench versus Claude Opus 5’s 50.3, 31.6 on Agents’ Last Exam (tying Opus 5), and 89.9 on Terminal Bench 2.1 versus Opus 5’s 89.1. These figures are as reported by SiliconANGLE; the article does not indicate independent replication.
• Licensing. VentureBeat states that MiMo-V2.6-Pro is open weight and MIT-licensed, and that users can download it for free off Hugging Face. It adds that Xiaomi is releasing more than 7,000 RL task environments and MiMo-V2.6-Distill-Qwen-9B, a smaller model distilled from MiMo RL trajectories.
• Pricing. According to VentureBeat, Xiaomi charges $0.435 per million uncached input tokens and $0.87 per million output tokens for Pro, and $0.14 per million uncached input tokens and $0.28 per million output tokens for Flash. SiliconANGLE reports that Xiaomi says Pro costs roughly one-20th to one-60th as much as overseas models at comparable intelligence, and that the UltraSpeed variant delivers up to 20 times faster output than Pro at the same quality.

What We Don’t Know

• Whether third parties will reproduce the AutomationBench, Agents’ Last Exam and Terminal Bench 2.1 results reported for the release.
• How the models perform outside the benchmarks cited, in production workloads.
• The Xiaomi documentation page does not itself specify license terms, parameter counts or pricing for the V2.6 series, so those details rest on the press coverage cited above.

Analysis

The release pairs a leading open-weight ranking with permissive licensing and low per-token pricing, a combination that gives developers who self-host or use hosted APIs a lower-cost alternative to closed frontier models. Whether the cost claim holds will depend on independent evaluations of the released weights.