VibeMathedMath problems solved with AI
← All frontiers
FrontierTheoretical computer sciencesignificance 55

Exponent of matrix multiplication

The exponent ω\omega: the smallest real number such that two n×nn \times n matrices can be multiplied in O(nω+ε)O(n^{\omega + \varepsilon}) operations for every ε>0\varepsilon > 0.

Current best · lower is better
2.3711772.371177
AlphaEvolve with Dupont, Eisenberger, Kozlovskii, Mehrabian, Ruiz, See, Zhou, Alman, Vassilevska Williams and Balog, 17 Aug 2026 · entry
steps
16
by AI
1
since
1969
1970198019902000201020202.42.52.62.72.81969: 2.8074 (Strassen)1978: 2.796 (Pan)1979: 2.780 (Bini, Capovani, Lotti and Romani)1981: 2.522 (Schönhage)1981: 2.517 (Romani)1981: 2.496 (Coppersmith and Winograd)1986: 2.479 (Strassen)1990: 2.3755 (Coppersmith and Winograd)2010: 2.3737 (Stothers)2012: 2.3729 (Vassilevska Williams)2014: 2.3728639 (Le Gall)2020: 2.3728596 (Alman and Vassilevska Williams)2022: 2.371866 (Duan, Wu and Zhou)2024: 2.371552 (Vassilevska Williams, Xu, Xu and Zhou)2024: 2.371339 (Alman, Duan, Vassilevska Williams, Xu, Xu and Zhou)2026-08-17: 2.371177 (AlphaEvolve with Dupont, Eisenberger, Kozlovskii, Mehrabian, Ruiz, See, Zhou, Alman, Vassilevska Williams and Balog)

The line is the frontier over time. Filled dots are steps that moved it; muted dots are results that did not. Orange dots are catalog entries, results with AI in the loop. Hollow dots are candidates under review and never move the line. Grey dots along the bottom edge are results from before the quantity had a number, placed there because they have no value on this axis. Dots that would overlap are nudged sideways a few pixels. Hover a dot for its value and attribution.

About this frontier

Trivially 2ω32 \le \omega \le 3, and the conjecture is ω=2\omega = 2. Every step since Strassen in 1969 has been an upper bound, and since Coppersmith and Winograd in 1990 every step has come from analysing higher powers of one tensor with the laser method. The record moves in the fourth decimal place and each move is a paper.

Every step, newest first

DateValueWhoModelStatusSource
17 Aug 20262.3711772.371177bestAlphaEvolve with Dupont, Eisenberger, Kozlovskii, Mehrabian, Ruiz, See, Zhou, Alman, Vassilevska Williams and BalogAlphaEvolveAI step
unreviewed
entry
20242.3713392.371339Alman, Duan, Vassilevska Williams, Xu, Xu and Zhouhistoricalsource ↗
20242.3715522.371552Vassilevska Williams, Xu, Xu and Zhouhistoricalsource ↗
20222.3718662.371866Duan, Wu and Zhouhistoricalsource ↗
20202.37285962.3728596Alman and Vassilevska Williamshistoricalsource ↗
20142.37286392.3728639Le Gallhistoricalsource ↗
20122.37292.3729Vassilevska Williamshistoricalsource ↗
20102.37372.3737Stothershistoricalsource ↗
19902.37552.3755
Stood for twenty years.
Coppersmith and Winogradhistoricalsource ↗
19862.4792.479Strassenhistoricalsource ↗
19812.4962.496Coppersmith and Winogradhistoricalsource ↗
19812.5172.517Romanihistoricalsource ↗
19812.5222.522Schönhagehistoricalsource ↗
19792.7802.780Bini, Capovani, Lotti and Romanihistoricalsource ↗
19782.7962.796Panhistoricalsource ↗
19692.80742.8074Strassenhistoricalsource ↗

Historical rows are the timeline table in the Wikipedia article on the computational complexity of matrix multiplication, which cites each paper; years are publication years as given there. Where two bounds share a year (1981) the table's order is kept.

Changelog1 change
  • Curatoradded this entry

Discussion