Tencent Releases Hyra-1.0, a Research Agent Capable of Recursive Self-Improvement
Tencent Hunyuan has released Hyra-1.0, its first research agent designed for recursive self-improvement in performance-oriented scientific discovery and engineering tasks. The agent is built to learn and evolve within real AI R&D pipelines, natural science research, and industrial environments. This release marks a significant step in the development of recursive self-improvement (RSI) agents, pushing AI capabilities beyond closed benchmarks into open-ended scientific discovery. It aligns with a growing industry trend toward automated AI research, joining efforts like Google DeepMind's AlphaEvolve. Hyra operates on a lightweight framework with a broad action space, continuously running a loop to propose better solutions based on historical logs, evaluator feedback, and source code. In a practical test, it successfully designed 3D structures and rendered 3D models from a 2D reference image, outperforming Claude Code's goal mode in visual similarity and human aesthetics.
## BACKGROUND
Recursive self-improvement (RSI) is a process where AI systems autonomously optimize their own code or algorithms to enhance their capabilities. Recently, evolutionary coding agents like Google DeepMind's AlphaEvolve have demonstrated the potential of this approach by optimizing complex algorithms for scientific fields such as quantum physics and genomics.