IBM researchers said their open-source OpenEvolve framework uncovered 465 new quantum error-correction codes, a result aimed at speeding a search process long constrained by huge algebraic design spaces.
The system uses large language models to propose promising algebraic expressions and an evolutionary loop to test and refine them, targeting bivariate bicycle codes on IBM’s fault-tolerant quantum roadmap.
One standout candidate, [[288,50,8]], lifts the logical-qubit count to 50 from the previous record of 16 in that code family, though with relatively low error-tolerance distance.
Other finds include a compact [[72,4,8]] code that could be easier to implement on near-term hardware and more balanced candidates such as [[288,16,12]] and [[360,12,≤24]] that compare with IBM’s [[144,12,12]] gross code.
IBM has released OpenEvolve on GitHub, while noting the AI-generated codes still need further evaluation on real physical quantum architectures.
AI just found 465 new quantum codes. How much faster will it now break our internet security?
As AI accelerates quantum discovery, what is the new role for human scientists in the lab?
AI-Powered QEC Code Discovery: IBM OpenEvolve and the Race to Fault-Tolerant Quantum Computing (QLOPS Benchmark in Focus)
Overview
In June 2026, IBM introduced OpenEvolve, an AI-driven framework designed to speed up the discovery of Quantum Error Correction (QEC) codes, which are vital for protecting quantum information from errors. The research team tested OpenEvolve by focusing on bivariate bicycle (BB) codes, a key type of quantum low-density parity check (qLDPC) code that supports IBM’s goal of achieving fault-tolerant quantum computing. This breakthrough marks a pivotal moment for building more robust quantum systems, as OpenEvolve enables faster and more efficient identification of error correction codes essential for reliable quantum computers.