Hybrid Open-Ended Tri-Evolution Makes Better Deep Researcher

2026-06-10

AI Summary

Current AI agents are limited in open-ended, deep research tasks due to their static parametric capabilities, and existing evolutionary approaches struggle to adapt to research scenarios without ground-truth answers. This work proposes the Hybrid Open-ended Tripartite Evolutionary framework (HOTE), which introduces for the first time a synergistic co-evolution mechanism among three distinct modules: proposer, solver, and critic. By integrating hybrid-mode reinforcement learning, web-scale knowledge utilization, and long-context training strategies, HOTE enables agents to autonomously conduct in-depth research in open environments. Evaluated on three long-form research benchmarks, HOTE with an 8B-parameter model outperforms state-of-the-art static open-source models ranging from 8B to 32B parameters as well as advanced training methods, while achieving higher training efficiency.

Abstract

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.

Problem

• open-ended research
• agent evolution
• deep research
• artificial general intelligence
• autonomous agents

Innovation

• Hybrid Open-Ended Tri-Evolution
• deep research
• agent evolution
• reinforcement learning
• autonomous AI agents