Key Notes
- Xiaomi released MiMo-V2.6-Pro and MiMo-V2.6-Flash under an MIT license, with Pro scoring 46 on Artificial Analysis' Intelligence Index to become the top rated open weight model.
- The 1.02 trillion parameter Pro model and smaller 310 billion parameter Flash model kept pricing flat from the previous generation, at 0.435 and 0.14 dollars per million input tokens respectively.
- Xiaomi also broadcast parts of the reinforcement learning process online and demonstrated the models generating 3D scenes, working inside Blender and Figma, and composing an orchestral score converted to MIDI.
Xiaomi has released MiMo-V2.6-Pro and MiMo-V2.6-Flash, two open weight models the company says advance through scaled reinforcement learning, with the flagship Pro model becoming the highest scoring open weight model on Artificial Analysis’ Intelligence Index.
Topping the Open Weight Leaderboard
MiMo-V2.6-Pro scored 46 points on Artificial Analysis’ Intelligence Index, according to VentureBeat, putting it ahead of Z.ai’s GLM-5.3 and Moonshot AI’s Kimi K3, which both sit at 44 points, and well above Xiaomi’s own previous generation model, MiMo-V2.5-Pro, which scored 26 on the same index. The score ties Grok 4.7, which xAI released the same day, and sits sixth overall among both open and proprietary models, with every model ranked above it kept closed.
Xiaomi says Pro performs on par with Anthropic’s Claude Opus 5 and OpenAI’s GPT-5.6 Sol across most agent benchmarks, though independent analysis from Kingy AI found the model still trails Claude Opus 5 on 10 of 14 shared evaluations, with the widest gaps on cybersecurity focused tests such as ExploitBench and on Terminal-Bench 4.0. Claude Opus 5 itself scores 51 on the Intelligence Index at its maximum effort setting, roughly 45 times more expensive per task than MiMo-V2.6-Pro.
A Trillion Parameter Model Built for Efficiency
MiMo-V2.6-Pro is a sparse mixture of experts model with 1.02 trillion total parameters, of which only 42 billion activate for any given inference, according to OfficeChai. The smaller Flash variant carries roughly 310 billion total parameters with about 15 billion active. Xiaomi’s team, led by former DeepSeek researcher Fuli Luo, says the architecture incorporates hybrid attention mechanisms and a multi token prediction decoder aimed at keeping inference fast despite the model’s overall size.
Pricing is unchanged from the prior generation. Flash is priced at 0.14 dollars per million input tokens and 0.28 dollars per million output tokens, while Pro costs 0.435 dollars and 0.87 dollars respectively, figures that put MiMo-V2.6-Pro’s cost per Intelligence Index task at roughly 0.13 dollars, among the cheapest of any model near the top of the leaderboard. Xiaomi also offers a Pro UltraSpeed variant that the company says delivers up to 20 times faster output for users who need extreme generation speed, at a significant premium over the standard Pro pricing.
Built In Public, at a Fraction of Frontier Lab Budgets
According to Xiaomi’s own account of the training run, reported by AI Weekly, the Pro model completed 30 large reinforcement learning steps covering roughly 750,000 trajectories in under six days, at a total cost of approximately 2.62 million dollars, with the smaller Flash model trained for about 850,000 dollars. Xiaomi says it assigned several dozen people to the reinforcement learning effort, a team size far smaller than those typically associated with frontier model training at US labs.
In demonstrations tied to the release, Xiaomi says the models can generate interactive 3D scenes from text, images, and video, operate inside design tools such as Blender and Figma, and assemble interface mockups and presentations using external tools and agents. The company also says Pro wrote an original orchestral score and converted it into a MIDI file without additional human editing, though these demonstrations reflect Xiaomi’s own framing of the model’s capabilities rather than independently reproduced results.
Part of a Crowded Open Weight Field
MiMo-V2.6 arrives amid a wave of open weight releases from Chinese AI labs, including Moonshot AI’s Kimi K3 and Alibaba’s Qwen3.8-Omni-Flash, all competing on a mix of raw benchmark performance and cost per task rather than on capability alone. Xiaomi is not traditionally grouped among China’s best known frontier labs, making its jump to the top of the open weight rankings notable within that field.
Whether MiMo-V2.6-Pro holds its position will depend on how quickly the leaderboard moves. Both Meta and OpenAI are reported to have models in testing that could outscore Xiaomi’s 46 point result once Artificial Analysis next updates its index, which would turn Xiaomi’s current claim to the top of the open weight rankings into a narrower, shorter lived one.
Disclaimer: AIstify is an independent media brand owned and operated by NuvexMedia LLC, publishing news, research, and insights on artificial intelligence, emerging technologies, automation, and related industries. NuvexMedia LLC invests in and collaborates with companies across the AI, technology, software, and digital innovation sectors. These relationships do not influence AIstify’s editorial coverage, and the publication maintains full editorial independence to provide accurate, timely, and objective information. © 2026 NuvexMedia LLC. All rights reserved. This content is for informational purposes only and should not be considered legal, tax, investment, financial, or other professional advice.
