Google’s Willow chip: magic-state cultivation cuts error by a factor of 40

**Google Quantum AI reported experimental magic-state cultivation on its Willow superconducting processor in a Science First Release dated Oct. 15, 2026 (DOI 10.1126/science.aee7961). The peer-reviewed abstract states: “Cultivation reduces the error by a factor of 40, with a state fidelity of at least 0.9999(1) (retaining 8% of attempts).”**

What the paper says

The article, “Magic state cultivation on a superconducting quantum processor,” was received Dec. 15, 2025, and accepted Oct. 1, 2026. Corresponding authors include E. Rosenfeld and K.J. Satzinger of Google.

According to the abstract, the team implemented cultivation, including code-switching into a surface code, and developed a fault-tolerant measurement protocol to bound magic-state fidelity. The work addresses the high resource cost of non-Clifford operations that most error-correction architectures need for a universal gate set.

The abstract’s keep rate — retaining 8% of attempts — means about 92% of attempts were discarded under postselection.

Google’s same-day blog

In a Google Quantum AI blog post the same day, Emma Rosenfeld and Craig Gidney wrote that cultivation on Willow reduced errors “about 40x compared to without it” and produced “T-states with a record logical fidelity of 99.99%”. They said the team grafted purified states from a distance-3 color code to a distance-5 surface code on hardware.

Google’s caveat, in the same post: “we still need to demonstrate magic state cultivation at a larger scale to make it realistic for practical quantum computing”.

Theory context

A 2024 theory paper by Gidney, Shutty and Jones, “Magic state cultivation: growing T states as cheap as CNOT gates” (arXiv:2409.17595), set out the protocol the Science experiment tests.

What this is not

Neither the Science abstract nor the Google blog states that a full quantum computer became “40x cheaper.” The measured claim is an error reduction by a factor of about 40 on cultivated versus uncultivated states, with heavy postselection. Viral “40x cheaper” framing overstates what was published.

What’s next / what’s unknown

Larger-scale demonstrations remain outstanding, by Google’s own account. Head-to-head experimental comparisons with distillation on the same Willow chip under identical noise were not reviewed for this draft.