New AI optimization framework beats Claude Code and Codex by 2.5x on the same compute budget

Researchers from Renmin University of China and Microsoft Research have introduced Arbor, a new AI optimization framework. Arbor is designed to improve the efficiency of AI model execution by optimising computational workflows.

According to the researchers, Arbor achieves a 2.5x performance improvement over existing tools such as Claude Code and Codex when operating under the same compute budget. The framework focuses on reducing redundancy in the optimization process.

The development highlights ongoing efforts to increase AI efficiency without additional hardware costs. Arbor is not yet publicly available, but the researchers have indicated plans to release the framework for further testing.