Research questionHow can research agents refine multi-constraint answers while keeping evidence verified over long horizons?A candidate answer can satisfy some constraints while violating others, and checking individual claims is often easier than discovering all required evidence. As research interactions grow, agents must retain which claims are supported and which remain unresolved. Latest papersRecent research connected to this question, newest first.AREX: Towards a Recursively Self-Improving Agent for Deep ResearchThe source describes AREX agents that gather evidence, construct provisional answers, audit constraints, and pursue unresolved claims through follow-up research. It reports training on verified synthetic tasks and research trajectories, results on BrowseComp, WideSearch, DeepSearchQA, Humanity’s Last Exam, and other reasoning and tool-use benchmarks, and experiments with dense 4B and 122B-A10B mixture-of-experts models.research paper · Sep 1, 2026