Tencent Hunyuan has released the first version, Hyra-1.0, an agent designed for performance-oriented research and engineering tasks that can recursively self-improve. It can learn and evolve in real AI development pipelines, natural science research, and industrial scenarios. The agent adheres to design principles of a lightweight framework and a broad action space, optimizing solutions through iterative exploration, and performs better in tasks such as generating 3D models from 2D reference images. Recursive self-improvement and automated research are at the forefront of AI, with several related achievements already emerging.
