Most of exact linear algebra on computers can be reduced to multiplying matrices. Faster methods could therefore speed up work on some fundamental problems in linear algebra and computer science.
What we know
The best known exponent is about 2.371; in 2026, an AI-assisted search lowered it slightly once more.
What blocks us
No one knows whether the ideal exponent of 2 is reachable.
Current leads
Algebraic methods, AI-driven algorithm search.
Sub-questions
Is the matrix multiplication exponent equal to 2, so that two n×n matrices can be multiplied in about n^(2+ε) steps for any ε > 0?
Can a new method get past the proven limits of the 'laser method', whose wide family of variants cannot go below an exponent of 2.3078?
How fast can rectangular (non-square) matrices of different shapes be multiplied?