Tencent Hunyuan Releases Hyra 2.0, Driving AI to Shift from Solving Research Challenges to Independently Developing Scientific Models and Tools

on November 2, according to official announcements from Tencent Hunyuan, Tencent Hunyuan officially launched the AI research system Hyra 2.0, shifting its focus from directly solving scientific problems to independently developing and upgrading reusable scientific models, research workflows, and software tools. The new system adopts an Overall, Mission, and Worker three-tier Agent architecture, supporting long-term task management, parallel experiments, independent verification, and the accumulation of research experience.

Tencent Hunyuan revealed that Hyra 2.0 has independently trained the protein design model HyInv and the protein language model HyPLM-Flash. HyInv achieves a sequence generation speed 57 times faster than ProteinMPNN, while HyPLM-Flash contains 650 million parameters and delivers the fastest inference speed among 12 comparative models.

Additionally, Hyra 2.0 has made advances in fields such as quantum error correction, robotic structural design, and chip research and development, including increasing the geometric mean of SPEC CPU2006 scores per GHz for the open-source RISC-V processor Xiangshan by 5.79%. These related results will be progressively released for community verification and reuse.