Researchers achieve 99.99% logical fidelity for T-states on a quantum machine, although only 8% of cultivation runs produce usable magic states

Quantum computing researchers have demonstrated on hardware a technique that could make error-corrected quantum computing substantially less resource-intensive. In a study published 15 October 2026 in the journal Science, a team showed that “magic state cultivation” reduced errors in special quantum states by about 40 times compared with uncultivated states, while achieving a reported logical fidelity of 99.99% for T-states.

Magic states are consumable resources required for the non-Clifford operations that give quantum computers their advantage over classical machines. They are costly to produce, and some estimates suggest their manufacture could consume as much as 95% of a quantum computer’s capacity. Each state can be used only once, according to a backgrounder by the Scientific American.

The conventional approach, distillation, combines many imperfect states to yield a smaller number of higher-quality ones. Cultivation instead improves a single state: researchers encode a rough magic state across a small set of qubits, measure whether it is correct, and retain only successful candidates. Those states are then expanded across more qubits for stronger protection.

The magic state experiment used 54 of the Google Willow processor’s 105 physical qubits. Without cultivation, roughly four in 1,000 attempts produced a faulty state; with it, the error rate fell by about 40 times. The researchers also transferred states from a distance-3 color code to a distance-5 surface code, two different error-correction schemes, showing that purified states can survive that transition on the hardware.

Potential gamechanger?

Independent researchers have welcomed the result:

  • Isaac Kim of the University of California, Davis, told New Scientist that the approach “can make fault-tolerant quantum computing substantially cheaper”.
  • Dominik Hangleiter of ETH Zurich described it as a potential “gamechanger” for machines with hundreds of logical qubits.

There are important caveats. Because cultivation discards states that fail its quality checks, only about 8% of runs produced usable magic states. Zhenyu Cai of Imperial College London has said Willow’s error rates must improve further before the system can address practically useful problems.

The accepted states currently show about one error per 10,000 attempts. Breaking RSA-2048 encryption would require roughly one error in 10 billion, while molecular simulation would require one in a million to one in 100 million, according to a Scientific American report.

The researchers have acknowledged that cultivation has yet to be demonstrated at larger scale. “This is the first demonstration that the concepts work,” said lead researcher Emma Rosenfeld. “We’re not done”.