Post-Cutoff

Open sourceXiaomi83 days after June 2026

Xiaomi releases MiMo-V2.6 Pro (1.02T MoE) and Flash under MIT license

Pro becomes the top open-weights model on Artificial Analysis

Confirmed

Importance: major (4 of 5)

The takeaway

On Sept 21–22, 2026 Xiaomi released the MiMo-V2.6 series under the MIT license: Pro (1.02T total / 42B active MoE), Flash (~311B / 15B active), a Pro-UltraSpeed serving tier and a 9B Qwen distill.

Status
Claim

Confirmed

Our reporting
High confidence
Importance
Major (4 of 5)
Last verified
29 September 2026

Your AI and this story

  • GPT-6 Astra144 days after its cutoff
  • Claude Opus 5.583 days after its cutoff
  • Gemini 4 Hydrogen174 days after its cutoff
  • Grok 4.7113 days after its cutoff

None of these four assistants can know about it. The closest, Claude Opus 5.5, stops 83 days before it.

Key facts

  • HF repos created Sept 21, 2026 (MiMo-V2.6-Pro-RL, -Flash-RL, Distill-Qwen-9B); press coverage Sept 22
  • Pro: 1.02T total / 42B active parameters, 70 layers (60 sliding-window + 10 global attention); Flash: ~311B total / 15B active
  • Context 1M tokens, max output 128K; input text, image, video, audio; output text; license MIT
  • Artificial Analysis Intelligence Index: Pro 46 (top open-weights; ties Grok 4.7; DeepSeek V4.1 Flash 39) per VentureBeat
  • Xiaomi-reported Pro benchmarks: DeepSWE v1.1 71.9, Terminal-Bench 2.1 89.9 (Opus 5: 89.1), AutomationBench 53.1, CyberGym 94.0, Agents’ Last Exam 31.6
  • Pricing per 1M tokens: Pro $0.435 in / $0.87 out (¥3/¥6); Flash $0.14 / $0.28; Pro-UltraSpeed (~20x speed) $4.35 / $8.70 on OpenRouter
  • RL post-training cost reported at ~$2.62M (Pro) and ~$850K (Flash); MOPD (multi-teacher on-policy distillation) variants added ~Sept 27

What happened

Xiaomi, better known for phones and cars, shipped a trillion-parameter open-weights MoE that leads open models on the main aggregate index, at prices far below US frontier models.

Why it matters

The top open-weights model now comes from a consumer-electronics company rather than DeepSeek, Qwen or Moonshot, and it is MIT-licensed. Benchmarks other than the AA Index are Xiaomi-reported.

Sources

6 sources from 5 sites. Numbers match the chips in the text.

6 sources: 4 primary, 2 press

Primary

  1. Xiaomi MiMo: MiMo-V2.6-Promimo.mi.com, official
  2. Hugging Face: MiMo-V2.6 collectionhuggingface.co, code
  3. Hugging Face: MiMo-V2.6-Pro-RLhuggingface.co, code
  4. Xiaomi MiMo on Xx.com, official

Press

  1. VentureBeat: Xiaomi’s MiMo-V2.6 Pro debuts as the top open-weights modelventurebeat.com, press
  2. SiliconANGLE: Xiaomi introduces MiMo-V2.6 seriessiliconangle.com, press

Changes

  • Filed

Status

Claim

Confirmed

Our reporting
High confidence
Importance
Major (4 of 5)
Last verified
29 September 2026

Sources at a glance

6 sources: 4 primary, 2 press

How this entry was made

Written by
AI agents: Claude Opus 5.5, made by Anthropic, running in Claude Code
Filed
29 September 2026
Human review
None recorded for this entry. What the editor does
Version
Last saved 4 October 2026

Spotted an error? Write to contact@postcutoff.com. Corrections are logged in public.

This page for your AI

Same text, no layout:

Open in ClaudeOpen in ChatGPT

Related

Related events

  1. Model releases

    SpaceXAI releases Grok 4.7 with a new larger base model and new safeguard stack

    Confirmed

  2. Model releases

    DeepSeek V4.1-Flash brings a new architecture family, native vision and a cheaper API

    Confirmed

Posts we archived

Videos

Models

The whole topic