Kimi Researcher

Kimi Researcher is Moonshot AI's first agent product — a deep-research agent built on end-to-end reinforcement learning rather than a predefined workflow, under the team's "model as Agent" philosophy . It began internal testing on June 20, 2025, and rolled out gradually thereafter . For each query it autonomously plans its own execution — per its technical blog, averaging over 200 URLs searched and 70+ search queries per run, delivering a cited report of 10,000+ characters . It scored 26.9% on Humanity's Last Exam, per 高榕创投's reporting edging past Claude 4 Opus (10.7%) and Gemini 2.5 Pro (21.6%) and essentially tying OpenAI's Deep Research (26.6%) . Typical uses, per the BlueRun-published team piece, include due-diligence-style research, literature review, and structuring unfamiliar knowledge domains — the team framed it not as a search tool but as "an AI agent that genuinely knows how to do research," with end-to-end RL as the harder but necessary path .