Xiaomi Open-Sources MiMo-V2.6: Pro Tops the Artificial Analysis Open-Weight Board
Xiaomi has released and open-sourced the MiMo-V2.6 series with fully open weights: three tiers (Pro / Flash / Ultraspeed) plus a desktop client, all natively omni-modal. Industry digests report the Pro tier tops the Artificial Analysis open-weight leaderboard, ahead of Kimi K3 and Qwen3.8 Max.
Source: AI 日报（5x10.cn）
Launch
According to industry daily reports, Xiaomi officially released and open-sourced the MiMo-V2.6 series on September 22, 2026, with fully open weights — independent developers can self-deploy to cut costs further, no longer "having to watch Llama and Mistral from abroad."
Three Tiers
- Pro: for complex projects
- Flash: for general scenarios
- Ultraspeed: for maximum-speed scenarios
- A desktop client ships alongside
The whole series is natively omni-modal; reports speculate special optimizations for edge-side inference efficiency (media speculation, not an official claim).
Leaderboard and Cost
- The Pro tier reportedly tops the Artificial Analysis open-weight leaderboard, scoring above Kimi K3 and Qwen3.8 Max
- API pricing unchanged, at roughly $0.13 per task — about one-third of comparable flagships
- Xiaomi describes it as "a key milestone on the path of recursive self-improvement"
Not Yet Disclosed
- Parameter count: not mentioned in reports
- Specific open-source license name: not mentioned
- Context length: not mentioned
September saw white-hot competition in open weights: DeepSeek V4.1 Flash (Sep 10) had just taken the open-weight crown when MiMo-V2.6 Pro (Sep 22) took it over. For our readers, the capability ceiling of locally deployable models keeps rising fast.