Google Reports a 40-Fold Error Cut in Making Quantum Computing's Magic States

Google Quantum AI says a technique called magic state cultivation, run on its Willow superconducting processor, cut errors in preparing a scarce class of quantum states by about 40-fold. The work was published Oct. 15, 2026, in Science.

Fault-tolerant quantum computers correct their own errors as they run, but the error-correcting codes in use can handle only part of the set of operations a general-purpose program needs. The rest are supplied by feeding the machine specially prepared qubits called magic states, which are used up on each operation and cannot be reused. Google's post says some estimates put the share of a quantum computer's capacity that would go to manufacturing them at up to 95%, leaving little room for the computation itself. So the cost of making them bears on how large a useful machine would have to be.

Google says the 40-fold figure is measured against states prepared without cultivation, and reports a fidelity of at least 99.99% for the states that came through, with 8% of attempts retained. Those are the company's own measurements on its own hardware, and the paper carrying them was peer reviewed by Science. Google says cultivation still has to be shown at a larger scale before it is realistic for practical quantum computing.

Cultivation works as a filter. A noisy state is injected into a small region of the chip, a series of coordinated measurements checks it for errors, and flawed attempts are thrown away, leaving a purer state that can then be grown into a larger encoding. The method was set out in 2024 by Google researchers Craig Gidney, Noah Shutty and Cody Jones, and the Science paper reports a test of it on hardware.

The team also switched between two kinds of error-correcting code during the work. States prepared in a color code, a newer scheme suited to cultivation, were grafted into a surface code, the more established of the two, and Google says they survived the transfer.