Hybrid Open-Ended Tri-Evolution Makes Better Deep Researcher
For AI agent research, this work bridges deep research and agent evolution for open-ended tasks, but the gains are incremental as it combines existing reinforcement learning and modular evolution techniques.
The paper proposes the Hybrid Open-Ended Tri-Evolution (HOTE) framework to improve deep research agents by enabling collaborative evolution of proposer, solver, and judge modules via hybrid-mode reinforcement learning. The 8B model trained with HOTE outperforms static open 8-32B models and state-of-the-art deep research training methods on three long-form benchmarks with less time overhead.
Deep research and agent evolution serve as de-facto tasks for AI agents in real-world applications toward artificial general intelligence. The former enables autonomous retrieval and integration of information in open-ended environments to tackle open-ended research tasks, yet it is constrained by the static parametric deep research capabilities of agent systems. The latter allows agents to autonomously interact with the environment to gain experiences that evolve model capabilities. However, its effectiveness has been widely validated only on verifiable tasks with standard answers, leaving a gap with open-ended research tasks. To bridge these two critical tasks, we propose the Hybrid Open-Ended Tri-Evolution (HOTE) framework, which leverages hybrid-mode reinforcement learning to facilitate the collaborative evolution of a proposer, solver and judge based on web-scale knowledge, moving toward autonomous evolving agents in open-ended tasks and environments. Extensive experiments on three long-form deep research benchmarks demonstrate that the 8B model trained via HOTE surpasses the strongest static open 8-32B models as well as those trained by state-of-the-art deep research training methods with less time overhead, and further verify that the evolution of all three modules in HOTE is indispensable.