Google DeepMind has moved AlphaEvolve, its Gemini-powered evolutionary algorithm agent, from a research project into broad real-world deployment, one year after the tool was first unveiled as a system for automated algorithm discovery.
AlphaEvolve pairs the creative problem-solving capabilities of Google's Gemini large language models with evolutionary computation, iteratively generating and refining candidate algorithms for hard computational problems.
The system uses Gemini Flash to generate a wide range of rapid code mutations and Gemini Pro to refine the most promising solutions, scoring each candidate against objective evaluation metrics before feeding the best results back into the next evolutionary cycle.
In its first year, AlphaEvolve advanced decades-old mathematical problems, including improving upon a matrix multiplication algorithm that had stood since 1969, and working with UCLA mathematician Terence Tao to solve open problems posed by Paul Erdős.
Google said the tool is now tackling challenges well beyond pure mathematics.
In earth sciences, AlphaEvolve improved the overall accuracy of natural disaster risk prediction by 5% across 20 categories, including wildfires, floods and tornadoes.
In quantum physics, it suggested quantum circuits with ten times lower error than conventionally optimised baselines, enabling complex molecular simulations to run on Google's Willow quantum processor.
Applied to power grid optimisation, AlphaEvolve increased the feasibility of solutions for the AC Optimal Power Flow problem from 14% to more than 88%.
Related reading
- OpenAI breaks free from Microsoft exclusivity in rewritten partnership deal
- AI platform Von targets sales teams with 'intelligence layer' for revenue operations
- Anthropic launches Claude Design in challenge to Figma and Adobe
Within Google's own infrastructure, the agent recovered 0.7% of global compute resources by improving data centre scheduling, delivered a 23% speedup in a key Gemini training kernel and contributed circuit-level optimisations to an upcoming tensor processing unit.
AlphaEvolve is now available in private preview through Google Cloud, where customers are using it to improve machine learning models, accelerate drug discovery, optimise supply chains and refine warehouse design.
The recap
- AlphaEvolve moves from research into practical problem solving
- Powered by Gemini and an evolutionary algorithm agent
- Company says it is now tackling major global challenges
