腾讯混元发布 Hy-Memory-30B-A3B，专攻智能体长期记忆写入
腾讯混元做了个只管写记忆的 3B 激活小模型，记忆基准分还超过了 Claude Opus 5，做智能体记忆的可以看看。
腾讯 Hunyuan 训练了 Hy-Memory-30B-A3B，专门决定智能体存什么、存多精确、以及新证据如何更新已有记忆。在四个评估框架（含 held-out 的 EverMemOS）的 20% 基准子样本上，它以 55.49% 的 macro 平均分排名第一，超过 Claude Opus 5 的 54.62%。该模型可运行在 Mem0、MemOS 和 Hy-Memory 三套框架中。训练分两阶段：先用 on-policy distillation 学习各框架的写入协议，再用 RL 奖励能让后续问题被答对的写入行为，并对存储过多施以惩罚。
A 3B-active model can now write long-term memory as well as Claude Opus 5.
Tencent Hunyuan trained Hy-Memory-30B-A3B for one job. It decides what an agent stores, how precisely, and how new evidence updates what is already there.
The same model runs inside Mem0, MemOS, and Hy-Memory. On fixed 20% benchmark subsamples, it ranks first on all four evaluated harnesses, including held-out EverMemOS, with a 55.49% macro average against 54.62% for Opus 5.
Training has two stages. On-policy distillation teaches each framework's write protocol. RL then rewards the writer when later questions can be answered from what it stored, with a penalty for storing too much.
https://t.co/If1TULtR0y
Chat with Paper: https://t.co/0GxMJwHTFn
