RSIAgent (arXiv:2609.15364): training-free multi-agent system with environment memory and claims vs GPT-6
RSIAgent (arXiv:2609.15364, 14 Sep 2026) proposes a training-free curriculum/actor/verifier with environment memory; on OSWorld-v2 and Agent's Last Exam the paper claims Kimi-K3 and GLM-5.3 beat some closed models including GPT-6—as a claim, not consensus.