AlphaEvolve: How the Gemini-Based Coding Agent Is Expanding Its Impact Across Multiple Fields

AlphaEvolve is a Gemini-based coding agent that is boosting performance across multiple fields through automated search and optimization.

• In life sciences, it improved Google Research's DeepConsensus, reducing DNA variant detection errors by 30%, enabling PacBio's genome analysis to be more accurate and lower cost.
• In power grid optimization, it raised the feasible solution discovery rate for AC Optimal Power Flow problems from 14% to over 88%.
• Through Earth AI model optimization, it improved average prediction accuracy across 20 disaster categories by 5%.
• In fundamental research, it helped Terence Tao solve Erdős problems, and also improved lower-bound records for the Traveling Salesman Problem and Ramsey Numbers.
• In quantum computing, it discovered circuits for Google's Willow chip with 10x lower error than before, contributing to experimental demonstrations.

AlphaEvolve has now moved beyond the experimental stage to become a core tool in Google's infrastructure. It has contributed to next-generation TPU design, improved cache replacement policies, a 20% reduction in write amplification for Google Spanner, and nearly 9% savings in software storage space through compiler optimization.

Commercialization is also underway with Google Cloud. Klarna doubled (2x) transformer training speed, and Substrate significantly improved computational lithography simulation speed. FM Logistic improved routing efficiency by 10.4%, cutting over 15,000km of travel distance annually.