Tencent Hunyuan released Hyra 2.0 on November 2, shifting the focus of automated research from directly solving problems to letting general agents develop and upgrade scientific models, workflows and software tools.
The system adopts a three-layer Overall-Mission-Worker architecture and connects to real R&D environments such as WeCom, compute platforms and code repositories. Results are openly published on GitHub under the Apache 2.0 license.
Hyra 2.0 has already produced a batch of tools, including the HyInv protein inverse-folding model (about 57 times faster than ProteinMPNN), the HyQEC quantum error-correction code (roughly 48% ahead on coding efficiency) and improvements to mathematical proofs and Xiangshan RISC-V processor optimization (a 5.79% per-GHz gain).
The release reflects a broader Chinese push to move AI from "cramming for exams" toward "building tools," embedding agents into the scientific discovery pipeline.
Source: China News Report (compiled from Daily AI Morning, Digital China Daily Brief).
