Author: AlphaEvolve team · Source: https://deepmind.google/blog/alphaevolve-a-gemini-powered-coding-agent-for-designing-advanced-algorithms/
Published: May 14, 2025 · Fetched: 2026-09-05
Note: body below is original English text extracted from the live page. Do not treat this file as a translation.
Today, we're announcing AlphaEvolve, an evolutionary coding agent powered by large language models for general-purpose algorithm discovery and optimization. AlphaEvolve pairs the creative problem-solving capabilities of our Gemini models with automated evaluators that verify answers, and uses an evolutionary framework to improve upon the most promising ideas.
AlphaEvolve enhanced the efficiency of Google's data centers, chip design and AI training processes — including training the large language models underlying AlphaEvolve itself. It has also helped design faster matrix multiplication algorithms and find new solutions to open mathematical problems.
Designing better algorithms with large language models
In 2023, we showed for the first time that large language models can generate functions written in computer code to help discover new and provably correct knowledge on an open scientific problem. AlphaEvolve is an agent that can go beyond single function discovery to evolve entire codebases and develop much more complex algorithms.
AlphaEvolve leverages an ensemble of state-of-the-art large language models: our fastest and most efficient model, Gemini Flash, maximizes the breadth of ideas explored, while our most powerful model, Gemini Pro, provides critical depth with insightful suggestions. Together, these models propose computer programs that implement algorithmic solutions as code.
Diagram showing how the prompt sampler first assembles a prompt for the language models, which then generate new programs. These programs are evaluated by evaluators and stored in the programs database. This database implements an evolutionary algorithm that determines which programs will be used for future prompts.
AlphaEvolve verifies, runs and scores the proposed programs using automated evaluation metrics. These metrics provide an objective, quantifiable assessment of each solution's accuracy and quality. This makes AlphaEvolve particularly helpful in a broad range of domains where progress can be clearly and systematically measured, like in math and computer science.
Impact highlights (as stated by DeepMind)
- Data center scheduling: a heuristic now in production for over a year continuously recovers, on average, 0.7% of Google's worldwide compute resources.
- Hardware design: a Verilog rewrite that removed unnecessary bits in a key arithmetic circuit for matrix multiplication; proposal must pass robust verification; integrated into an upcoming TPU.
- AI training/inference: sped up a vital kernel in Gemini's architecture by 23%, leading to a 1% reduction in Gemini's training time; up to 32.5% speedup for FlashAttention kernel GPU instructions.
- Matrix multiplication: found an algorithm to multiply 4×4 complex-valued matrices using 48 scalar multiplications, improving upon Strassen's 1969 algorithm in this setting.
- Open math problems: applied to over 50 open problems; in roughly 75% of cases rediscovered state-of-the-art solutions; in 20% of cases improved previously best known solutions (e.g., kissing number lower bound in 11 dimensions with 593 outer spheres).
Isolation note (for KB indexing)
Candidate = algorithmic programs / code under evolution. Optimizer = Gemini Flash+Pro ensemble + evolutionary prompt sampler / programs database. Eval & control = automated evaluators with objective metrics; proposals must pass verification before production integration. The generator and evaluator roles are not the objects being rewritten by the evolutionary loop in the same sense as the candidate algorithms.