Gemini-powered evolutionary coding agent for scientific and algorithmic discovery. Found the first improvement to Strassen's matrix multiplication in 56 years (4×4 complex matrices in 48 multiplications).

Also optimized Google data center scheduling, chip design, and LLM training kernels. Represents a new paradigm of AI-driven algorithm discovery where LLMs generate and evolve code solutions. By Novikov et al. (DeepMind).

In August 2026 a DeepMind paper with academic co-authors, including Josh Alman and Virginia Vassilevska Williams, used AlphaEvolve to rewrite the optimization code behind the laser method's combination-loss analysis. It lowered the upper bound on the matrix multiplication exponent to ω < 2.371177, from 2.371339, with rational-arithmetic certificates making the bound verifiable (arXiv 2608.16884). On 2026-11-04 DeepMind opened the google-deepmind/alphaevolve repository. It contains a minimal local implementation of AlphaEvolve, the evolved ω optimization program with its verifier and a Colab notebook, and the initial programs and results for the earlier mathematical problems.

Paper

Citations 6
scientificcodingagents

Related