🤔深度剖析：小米是如何把开源模型的游戏规则悄悄改变的？！
This is a dev post classified by Jev as Open source & devtools business (a launch), kept by the Dev Radar because it carries real work, not commentary.
🤔深度剖析：小米是如何把开源模型的游戏规则悄悄改变的？！ MiMo-V2.6-Pro 46.32 登顶 AA 开源智力榜，前代只有 26；最终 RL 开跑不到 7 天，模型 + 论文一起交卷，还要公开 ~7k RL 数据和整套框架！ MiMo V2.6 Pro & Flash 的关键秘诀： 1️⃣不拆领域，一次混合 RL 打穿代码 / Agent / 视觉 / 安全 2️⃣每步 1568×16 全异步 GRPO，3.5B+ tokens，最长 1M context 3️⃣奖励也 scale：组内对比打分 GRS + 优势重分配 GAR 4️⃣冻结 MoE 路由 + 多层防 reward hacking，万亿级才能稳 5️⃣预训练 Adam，中训 Muon+row-norm（Muown）硬撑大 batch 6️⃣Hybrid SWA/GA + MTP，Pro 1.02T/42B，Flash 310B/15B 7️⃣6 天、75 万轨迹：Flash $85 万、Pro $262 万，DeepSWE 直接 +14~17 分 8️⃣权重、环境、RL 代码、技术报告全部公开，走 RSI 而不是藏配方 智力上涨，价格还是 V2.5。这才叫把帕累托前沿往外推。 查看论文：https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-RL/b
Posted by SuSu_酥酥👅 (16.2k followers) 10 h ago · 52 likes · 13.9k views · view the original post on X. Kept by the Dev Radar as Open source & devtools business. Tools mentioned: mimo-v2.6-pro-rl.
More dev work like this
- What’s going on here lol what’s the point of these ai reply bots — @lu__jasper
- Sir, China Just drop another Open-model that "Claude Opus 5, GPT-5.6 Sol" level… — @0x0SojalSec
- What's my next big project? in SF cooking up http://opensourceaiweek.com with the… — @ekang426
- 😱卧槽！小米MiMo-V2.6 Pro 重新定义了斩杀区！ — @NFT_Chen
- omg xiaomi cooked hard with mimo 2.6 — @notjazii
- Xiaomi is contributing a LOT to open research with this release. — @lu__jasper
- 💥小米AI模型 MiMo-V2.6 Pro & Flash 深夜突袭！杀疯了！ — @NFT_Chen
- the most insane part, they will release ~7k RL training data and the framework leading… — @eliebakouch
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. 20.7k posts from 4.9k X accounts over the last 21 days, 2.4k tools, 12 markets. Collected every 5 minutes, fully re-ranked every hour — last update 2026-09-22 08:21 UTC. Full method.