Tencent hunyuan unveils hyra 2.0 to foster reusable tools in AI model development

Tencent Hunyuan has released Hyra 2.0, an AI research system, shifting the focus of AI development from directly solving scientific problems to enabling general Agents to participate in the creation and enhancement of scientific models, research workflows, and software tools. This new paradigm aims to ensure that the outcomes of one research effort can be repeatedly utilized in subsequent studies, moving the scientific approach from mere problem-solving to tool creation. Building upon the success of Hyra 1.0, which previously helped an open-source model resolve a fifty-year-old open problem in additive combinatorics by validating the viability of providing AI with clearly defined evaluation standards and iterative attempts, Hyra 2.0 introduces the concept of intelligent compounding through reusable tool upgrades. The architecture of Hyra 2.0 is structured into three layers: Overall, Mission, and Worker. The Overall layer aligns goals with human researchers, the Mission layer manages a single research question and judges strategic choices, while the Worker layer independently reads code, conducts experiments, analyzes failures, and can challenge the Mission's decisions. The system also incorporates three distinct storage mechanisms—project status, layered Agent memory, and a shared knowledge base called Books—allowing transferable methods to be organized as skills for direct invocation in future tasks during project reviews.