1
MemGym: a Long-Horizon Memory Environment for LLM Agents
专为LLM智能体设计的长期记忆测试环境MemGym，填补长周期任务基准空白。
arXiv:2605.20833v1 Announce Type: new Abstract: Memory is a central capability for LLM agents operating across long-horizon tasks. Existing memory ben…