NeuralStackMIT researchers published a new quantum error correction code on July 18 2026. The code achieves 95 percent logical qubit fidelity using 72 physical qubits. Tests ran on a 256-qubit superconducting processor.
The method reduces overhead by 40 percent compared with surface codes. It corrects both bit-flip and phase-flip errors in a single round. The team used reinforcement learning to optimize the decoding graph.
Previous records stood at 89 percent fidelity in 2025. The paper appears in Nature Physics this week.
The work was funded by the National Science Foundation and Google Quantum AI. IBM provided additional hardware access for validation.
Why this matters
The breakthrough reduces the qubit count needed for fault-tolerant quantum computers. Companies can now plan smaller-scale systems for chemistry simulations by 2029.
Further improvements could enable early quantum advantage in drug discovery. MIT plans to release open-source decoding software in August 2026.