Tencent's Hunyuan team released Hyra 2.0 on November 2, an AI research system that moves its focus from using AI to crack scientific problems directly toward having general agents build and upgrade scientific models, research workflows and software tools so that one effort can be reused across later studies. The earlier Hyra 1.0, driven by an open-source model, solved a combinatorics problem that had been open for more than five decades, validating an approach of giving AI hard problems with clear evaluation criteria and letting it iterate. Hyra 2.0 splits research work into three layers: Overall, which aligns goals with human researchers; Mission, which owns a single research question over time and decides which routes to pursue; and Worker, which reads code, runs experiments, analyzes failures and can challenge the Mission's judgment. The system keeps project state, layered agent memory and a shared knowledge base called Books in separate stores, and can distill transferable methods into reusable skills during reviews.
Source: public reports and official announcements
广告

