Event ·Multi-source

Xiaomi 发布 MiMo-V2.6，将开源 RL 训练全程

Xiaomi 发布 MiMo-V2.6 Pro 和 Flash 两个版本，此前公开了强化学习（RL）训练阶段的进展，并计划将训练过程全部开源，供外部研究者继续探索。基准测试显示，MiMo-V2.6 Pro 超过 Qwen3.8 Max、GLM-5.3、Kimi K3 等开源模型，但仍落后于 Claude Opus 5、OpenAI GPT 6 Astra 等闭源模型。定价上，Flash 为 0.14/0.28 美元每百万 token，Pro 为 0.435/0.87 美元每百万 token，维持此前与 DeepSeek 竞争时的降价，而 DeepSeek 已经涨价。按 Artificial Analysis 指数，MiMo-V2.6 是运行成本低于 1 美元每任务组中性价比靠前的模型。

Current read

Xiaomi 把 RL 训练全开源，Pro 版跑赢 Qwen3.8 Max，价格比涨价后的 DeepSeek 更低。

11 sources64° AI heatLast updated 2026/9/22 01:30:57

Evidence

11 sources

Primary sourceBlognone

2026/9/22 01:30:57View source

Related source 1OpenRouter

2026/9/21 21:08:07View source

Related source 2shao__meng

2026/9/23 01:44:23View source

Related source 3IT之家

2026/9/20 03:19:25View source

Related source 4宝玉

2026/9/22 16:49:55View source

Related source 5Techsauce

2026/9/23 01:56:02View source

Related source 6AI Will

2026/9/20 08:54:26View source

Related source 7Genk

2026/9/20 11:01:00View source

Related source 8歸藏(guizang.ai)

2026/9/22 02:55:05View source

Related source 9Decoder

2026/9/22 16:31:07View source

Related source 10SuperTechFans

2026/9/22 23:15:45View source

Conflict check

Dedup data only shows that these sources discuss the same event, not that they agree. There is no structured supporting or refuting evidence yet, so no conflict is inferred automatically.

Open the story