Hybrid Open-Ended Tri-Evolution Makes Better Deep Researcher

In three linesHOTE (Hybrid Open-Ended Tri-Evolution) is a hybrid reinforcement learning framework for autonomous evolution of AI agents on open-ended research tasks. An 8B model trained via HOTE outperforms static 8-32B models and state-of-the-art deep research methods on three long-form research benchmarks.

Summary generated by Claude — human-verified