刚刚，小米Mimo正式发布并全面开源MiMo-V2.6。
This is a AI post classified by Jev as Frontier models (a model release), kept by the AI Radar because it carries real work, not commentary.
刚刚，小米Mimo正式发布并全面开源MiMo-V2.6。 这次有点不一样，MiMo-V2.6直接把一个词写进了发布主题里：RSI，递归自我改进。 简单讲，就是继续扩大强化学习的规模，让模型进入真实环境里不断做任务、拿反馈、继续学习，尝试让模型自己把自己越练越强。 这轮一共发布 MiMo-V2.6-Pro 和 Flash 两个原生全模态模型。 @XiaomiMiMo 官方称，MiMo-V2.6-Pro 在 Artificial Analysis Intelligence Index 拿到 46 分，超过 Kimi K3 和 Qwen3.8 Max，成为目前最强的开源模型（甚至超过了Grok4.7）。 在官方公布的 Agent Benchmark 里： Terminal Bench 2.1 做到 89.9，超过表中 Claude Opus 5 的 89.1 和 GPT-5.6 Sol 的 88.8； AutomationBench 达到 53.1； Agent’s Last Exam 达到 31.6； MiMo Visual Coding 达到 72.3。 更值得看的是这次模型的RL过程，本次RL是全程直播，在不到 6 天的 Live RL 训练里： Flash 花了约 85 万美元，Pro 花了约 262 万美元； 两个模型分别跑了 30 个训练 Step；
Posted by Max For AI (39.6k followers) 1 h ago · 14 likes · 4.1k views · view the original post on X. Kept by the AI Radar as Frontier models.
More AI work like this
- Grok 4.7终于发布了，延迟了10天，结果确是希望越大失望越大，Artifical Analysis位列第六 — @xueyu1125
- 小米推出mimo-2.6 pro和flash — @mylifcc
- Alibaba releases OmniVChat — @HuggingPapers
- Grok 4.7 just released and it's an EXCELLENT model — @AlexFinn
- MiniMax H3 Max by @fal takes the #1 spot on Video Editing by Design Arena with an Elo of… — @DesignArena
- 🎉 Congrats to @XiaomiMiMo on MiMo-V2.6: day-0 support in vLLM for both sizes. — @vllm_project
- Respect to Xiaomi for pacing the frontier using only 30 steps — @nrehiew_
- Grok 4.7 just made 4.6 look old. — @minchoi
Every post is read and classified by Jev (TypeSafe): what it is, which market it belongs to, and whether the link is a real tool. 40.7k posts from 5k X accounts over the last 14 days, 1.7k tools, 19 markets. Collected every 5 minutes, fully re-ranked every hour — last update 2026-09-22 01:04 UTC. Full method.