Tencent Hunyuan releases AI research system Hyra 2.0, shifting toward having Agents develop reusable tools
Tencent Hunyuan releases AI research system Hyra 2.0, with the R&D focus shifting from directly using AI to tackle scientific problems toward having general-purpose Agents participate in the development and upgrading of scientific models, research workflows, and software tools, so that the results can be reused repeatedly in subsequent research. Architecturally, Hyra 2.0 splits research work into three layers: Overall, Mission, and Worker; Workers can independently read code, run experiments, analyze the causes of failures, and question Mission's judgments. The system also designs three separate storage systems: project state, Agent layered memory, and a shared knowledge base called Books. During retrospectives, transferable methods can be organized into skills for use in subsequent tasks.
Why read this
AI score 60
The original text explains Hyra 2.0's shift from solving problems to building tools, along with its three-layer Agent architecture, which can be used to understand the division-of-labor design of research agents.
Topics
Body · AI translation
Tencent Hunyuan Releases Hyra 2.0: Building Reusable Tools for Model R&D
2026-11-02 21:08 IT Home - Yuanyang
IT Home, November 2 news: Today, Tencent Hunyuan released the AI research system Hyra 2.0, shifting its R&D focus from directly using AI to tackle scientific problems toward having general-purpose Agents participate in the R&D and upgrading of scientific models, research workflows, and software tools, so that the results of one R&D effort can be repeatedly used in subsequent research, driving the research paradigm from "solving problem sets" toward "building tools."
Previously, Hyra 1.0 drove open-source models to solve an open problem in additive combinatorics that had been unresolved for more than fifty years, validating the feasibility of the approach of "giving AI difficult problems with clear evaluation criteria and having it try repeatedly." Building on this, Hyra 2.0 further proposes that the gains brought by tool upgrades can be repeatedly reused by downstream tasks, thereby forming an intelligence compound interest.
IT Home notes that architecturally, Hyra 2.0 splits research work into three layers: Overall, Mission, and Worker. Overall is responsible for communicating with human researchers and aligning goals; Mission is responsible long-term for a single research problem and judges trade-offs in approach; and Worker independently reads code, conducts experiments, analyzes the causes of failure, and can question Mission's judgments. The system also designs three separate storage mechanisms: project state, Agent hierarchical memory, and a shared knowledge base called Books. During project retrospectives, transferable methods can be organized into skills for direct use by subsequent tasks.