- Start
- Sep 22, 202690% CONFIDENCEfrom the source
MiMo-V2.6-Pro and MiMo-V2.6-Flash Released
- Xiaomi released and open-sourced the MiMo-V2.6 series on 22 September 2026: MiMo-V2.6-Pro, "our most capable model to date", and MiMo-V2.6-Flash, which "strikes the best balance between intelligence, efficiency, and cost", plus a MiMo-V2.6-Pro-UltraSpeed mode with up to 20x faster output at the same quality[1][5]
- Pro scores 46.32 on the Artificial Analysis Intelligence Index, "surpassing Kimi K3 and Qwen3.8 Max to become the strongest open-source model to date", though SiliconANGLE notes it still trails Claude Fable 5.1 and GPT-6 Astra on aggregate measures[1][4][5]
- Xiaomi streamed the reinforcement-learning production run live: in under six days each model completed 30 RL steps over roughly 750k trajectories, at about $0.85M for Flash and $2.62M for Pro, lifting DeepSWE v1.1 from 48.8 to 65.68 (Flash) and from 58.4 to 72.57 (Pro)[1]
- The weights ship under the MIT license with the technical report, more than 7,000 RL training environments and the RL code, and a smaller MiMo-V2.6-Distill-Qwen-9B research checkpoint[1][4][6]
- Both models are live in AI Studio, MiMo Code, MiMo Desktop (which leaves early access with this release), the MiMo API Platform and OpenRouter, with API pricing unchanged from V2.5[1]
Notable features
- Natively omnimodal: text, image, video and audio in, text out, with a 1M-token context window; Flash allows up to 128K output tokens[2][6]
- Pro is a sparse mixture-of-experts model with 1.02T total and 42B activated parameters; Flash has 309B total and 15B activated, and both use a 681M-parameter vision encoder and a 308M-parameter audio tokenizer[5][6]
- Prices per million tokens: Flash $0.14 input and $0.28 output, Pro $0.435 and $0.87, Pro-UltraSpeed $4.35 and $8.70, with cache hits from $0.0028 and cache writes free for a limited time[1][2]
- Agentic results against Claude Opus 5: AutomationBench 53.1 against 50.3, Agents' Last Exam 31.6 (a tie) and Terminal Bench 2.1 89.9 against 89.1[5]
- Xiaomi's "Vibe World" demos have it build interactive game worlds, model 3D objects in Blender, control a Franka Panda robotic arm from camera feeds, compose an orchestral piece as MIDI and help complete a 6,000-line Lean 4 formalization of Li and Yorke's "Period Three Implies Chaos"[1]
References 683% CONFIDENCEOverall confidence: 83%How well the pin's source and references back up its dates.Weighted average of how firmly 6 references, the source included, support the pin's start and end times; a reference counts half as much for every 180 days older than the newestShow all pins at 75% confidence or better
The first entry is always the pin's source. Overall confidence is a weighted average of how firmly each reference supports the start and end times used above; a reference counts half as much for every 180 days older than the newest.
- [1]90%mimo.xiaomi.com/mimo-v2-6mimo.xiaomi.com· Posted Sep 28, 2026· Starts Sep 22, 2026 ✓· 17% of score
Xiaomi's announcement is dated "September 22nd, 2026" and begins "Today, we are releasing and open-sourcing the MiMo-V2.6 series"; SiliconANGLE's[5] 22 September report says Xiaomi "today announced the release and open-sourcing".
- [2]90%MiMo-V2.6-Flash - Xiaomi MiMomimo.mi.com· Added Sep 28, 2026· 17% of score
Xiaomi's[1] model page for Flash: text, image, video and audio input with text output, 1M-token context, 128K max output, and $0.14 per million input tokens (cache miss), $0.0028 on a cache hit and $0.28 per million output tokens.
- [3]70%Xiaomi MiMo - Wikipediaen.wikipedia.org· Added Sep 28, 2026· 17% of score
Lists MiMo-V2.6-Pro (1.02 trillion parameters, 42 billion active) and MiMo-V2.6-Flash (310B, 15 billion active) with a release date of 21 September 2026 - the UTC day the models went live, the evening before Xiaomi's[1] 22 September (China time) announcement.
- [4]75%Xiaomi Open-Sources MiMo-V2.6 With Pro, Flash, and 9B AI Models | eWeekeweek.com· Published Sep 23, 2026· 17% of score
eWeek adds the third checkpoint, MiMo-V2.6-Distill-Qwen-9B (a fine-tune of Alibaba's Qwen3.5-9B), the 7,000+ RL environments released with the weights, and that Artificial Analysis independently scores Pro at 46 on its Intelligence Index.
- [5]85%Xiaomi introduces Mimo-V2.6 series open-source AI model family - SiliconANGLEsiliconangle.com· Published Sep 22, 2026· Starts Sep 22, 2026· 16% of score
SiliconANGLE's same-day report: Xiaomi[1] "today announced the release and open-sourcing of its MiMo-V2.6 series"; gives Pro's agentic scores against Claude Opus 5 (AutomationBench 53.1 v 50.3, Terminal Bench 2.1 89.9 v 89.1) and notes it still trails Claude Fable 5.1 and GPT-6 Astra on aggregate measures.
Suggest a correction
Something missing or wrong? Say it in your own words: a link that backs this pin up, a different start or end date and why, or a fact it lacks or gets wrong. The AI checks it against this pin's sources, searches for better ones, and adds any page that backs you up. The pin's own sources still count most. A picture that shows something else, or shows it badly, is looked at too, and moved down or replaced.
