推文

@yueke · 2026-10-16 22:12

Memory is quietly becoming the real differentiator for agents — most evals only look at reasoning, but long-running tasks die on memory. An open 30B-A3B memory specialist plus a framework around it feels like the right granularity. Curious to see how it holds up in real workflows.

曝光 2 · 评论 0 · 点赞 0 · 书签 0