Tencent Hunyuan releases the research Agent Hyra 2.0, pushing "AI doing research" from problem-solving toward wilder goals—letting machines independently develop scientific models, algorithms, and tools that can be reused over the long term.

To support research that can last weeks at a time, Hyra 2.0 splits the work across three layers of Agents: Overall handles resource coordination, Mission handles planning, and Worker runs experiments. Task states are persistent, so it can keep other projects moving during training gaps; every conclusion must be backed by code and logs as evidence, anything in doubt is sent to an independent Agent for review, and proven experience is distilled into reusable Skills.

From proteins to chips and robotic hands, the scorecard is already out