Google Willow cuts magic state errors 40x in Science paper

Google's quantum team has shown, for the first time on real hardware, a cheaper way to make the special "magic states" that a useful quantum computer needs. The result, published in Science on October 15, 2026, cut errors in those states about 40 times on Google's Willow chip. Magic states may eat up to 95% of a future machine's capacity, so making them cheaper attacks one of the field's biggest costs.

Google's announcement, written by research scientist Emma Rosenfeld and quantum software engineer Craig Gidney, puts it simply: "We've shown that magic state cultivation works on Google Quantum AI hardware."

What a magic state is

A quantum computer works on qubits, which can hold blends of 0 and 1 at once. Qubits are fragile, so practical machines will use error correction. That means spreading one reliable "logical" qubit across many physical ones.

Error-corrected machines can do some operations cheaply. Others they cannot do on their own. For those, the machine prepares a magic state in advance and then uses it up to perform the operation, Google explains. Each one works only once. T-states are one type of magic state.

Making magic states is costly. Google's post says that "up to 95% of a quantum computer's entire capacity might need to be dedicated to making them." Quantum Computing Report says older methods, called distillation, were projected to use 90% to 95% of a machine's physical qubits. Google says the problem was first foreseen in 2004.

How cultivation works

Gidney, Noah Shutty and Cody Jones of Google Quantum AI proposed magic state cultivation in 2024, building on an earlier idea. The new paper is the first time it has run on a real chip.

The Quantum Insider describes it as a filter. The chip spreads a rough magic state across a small group of qubits. It then runs a series of checks, throws away any attempt that fails, and grows the ones that pass.

Here, the team grew a state from a small error-correcting code into a larger one. The two codes are known as a distance-3 color code and a distance-5 surface code. Quantum Computing Report calls it "the first physical hardware implementation of magic state cultivation."

The results

Measure	Result	Source
Error reduction	About 40x versus uncultivated states	Google
T-state logical fidelity	99.99%, a record per Google	Google
Attempts kept	8%	Quantum Computing Report
Qubits used	54 of Willow's 105	The Quantum Insider
Storage test	Errors about 7x higher than simulated	The Quantum Insider

A 99.99% fidelity means roughly one error in 10,000 states. The low 8% keep rate is the price of the filter: most attempts are thrown away.

What is still unproven

The authors are clear about the limits. "Of course, we still need to demonstrate magic state cultivation at a larger scale," Google's post says. It gives no timeline for a useful quantum computer.

The Quantum Insider reports that stored states decayed faster than simulations predicted, with an error rate about seven times higher. It traces most of that to leakage, where a qubit slips out of the two states it should hold. The simulations had left that effect out. The outlet calls the work "an important, incremental advance toward practical quantum computing."

Getting the hardware to cooperate took work too. Interesting Engineering reports that early results were off because of drift in electrical equipment and changes in lab temperature. Google fixed them with better controls and a new way of benchmarking.

What this means for developers

Nothing changes in your stack this year. Google gives no date for a useful machine, and the method still has to work at a much larger scale. Treat this as a cost cut on paper, not a product.

Watch three numbers in follow-up papers: the 8% keep rate, the error rate during storage, and the qubit count. A useful machine needs magic states made fast and in bulk. A filter that discards 92% of attempts must run quickly to keep up.

If you buy cloud quantum time or track the field for your company, ask vendors how they plan to supply magic states. That one line item may decide how large and how costly a useful machine has to be.

For security teams, the advice stays the same. Move toward post-quantum cryptography, encryption designed to resist quantum attacks, on your own schedule rather than in response to headlines. Tools are already shipping: OpenSSH 10.6 added a post-quantum signature type on October 6.

The full paper, "Magic state cultivation on a superconducting quantum processor," is in Science under DOI 10.1126/science.aee7961.