返回全部动态

小米MiMo-V2.6-Pro登顶开源模型榜,Anthropic指控其蒸馏Claude

原标题:Xiaomi's affordable flagship AI leads the open models, and Anthropic says Claude helped get it there

THE DECODER模型发布质量 75

AI 摘要

小米发布MiMo-V2.6系列开源模型,旗舰版MiMo-V2.6-Pro在Artificial Analysis智能指数上得分46,成为当前最强开源模型,且每百万输入/输出token价格仅为0.435/0.87美元,单任务成本约0.13美元。性能提升主要来自大规模扩展的强化学习,训练不到六天,Pro成本约262万美元。小米同时开源了RL工具包、训练框架及约7000个带自动评分器的训练任务。但Anthropic此前威胁情报报告点名小米等七家中国实验室,指控其通过Claude非法蒸馏数据训练自有模型。

以上摘要由 AI 生成,可能存在误差。事实请以原文为准。

正文节选

Xiaomi's affordable flagship AI leads the open models, and Anthropic says Claude helped get it there Key Points - Xiaomi's new MiMo-V2.6-Pro model leads current rankings of open AI models while costing far less than the competition. - Xiaomi achieved the performance jump through expanded reinforcement learning, and the company released its tools and training tasks openly. - At the same time, Anthropic accuses Xiaomi of improperly siphoning off data through the Claude model to train its own model


发布时间:2026-09-22 19:42
抓取时间:2026-09-22 20:34
来源机构:THE DECODER
阅读原文the-decoder.com