DeepMind's AlphaEvolve Sets New Matrix Multiplication Exponent Record
Google DeepMind announced that, together with academic collaborators and AlphaEvolve—a Gemini-powered coding agent—it has improved the upper bound on the matrix multiplication complexity exponent ω, pushing it from 2.371339 to below 2.371177, a new record. Matrix multiplication is the fundamental operation underpinning modern computing, including AI, and its theoretical optimal exponent ω has long been a famous open problem in complexity theory. The result was first announced on August 18 by DeepMind research lead Pushmeet Kohli, showcasing the potential of AI-assisted algorithm optimization for solving foundational math problems.
Confirmed
- DeepMind improved the upper bound on the matrix multiplication exponent from 2.371339 to ω < 2.371177, achieved by AlphaEvolve combined with modern optimization methods.
- The technical approach includes reformulating the optimization problem, improving the combinatorial loss analysis, and introducing the machine-learning-based algorithm AlphaEvolve (m4, @deepmind).
- AlphaEvolve is a Gemini-powered coding agent, and the research is a collaboration between DeepMind and academia (m2, @DrSingularity).
Why it matters
- Matrix multiplication is the core foundational operation of modern computing and AI, so any improvement in ω is symbolically significant in theory.
- As @PetarV93 noted, while the improvement here is small, it is a substantive breakthrough for AI agents in fundamental algorithm discovery, hinting at far greater potential for AI-assisted mathematical research.
2026-08-18 ~ 2026-08-19 · 6 related posts
Primary sources
- [source] DeepMind Improves Matrix Multiplication Exponent with AlphaEvolve — deepmind · 2026-08-18
- [source] DeepMind's AlphaEvolve sets new record for matrix multiplication exponent, ω < 2.371177 — pushmeet · 2026-08-18
4 near-duplicate retellings: PetarV_93 · yoavgo · Dr_Singularity · Dr_Singularity