Google's Willow Processor Wins the Magic Square Game 98.3% of the Time: What the Science Paper of 19 November 2026 Shows About Quantum Contextuality at 105 Qubits
What Google Quantum AI published in Science on 19 November 2026
Google Quantum AI reports in Science that its superconducting processors display quantum contextuality, the property that a measurement result cannot be fixed in advance independently of which compatible measurements accompany it. On the Willow processor two players sharing entangled qubits won the magic square game in 98.30 percent of 900,000 rounds, where the best classical strategy wins at most 88.89 percent. A state-independent test returned a score of 5.590 against a noncontextual bound of 4 and a quantum maximum of 6. A multiplayer parity game stayed above its classical bound with 71 players, and shallow circuits solved hidden-linear-function problems on grids of up to 105 qubits.
Which claims the result supports for buyers and policymakers
The paper shows that a nonclassical resource survives on large, noisy hardware and that games with a physical ceiling expose coherent errors that random-circuit tests average away. It makes no claim about speed, cost or a useful application, and the 105-qubit result is measured against a conjectured classical bound. For organizations reading quantum announcements, the more durable contribution is a quantum hardware benchmark whose target no vendor can move. This post reads the claims against the published account and explains why nothing in it changes post-quantum migration plans.
Google's Magic-State Cultivation Result in Science, 15 October 2026: What the Willow Experiment Tests in the Under-a-Million-Qubit RSA-2048 Estimate
What Google published in Science on 15 October 2026
Google Quantum AI reports in Science that magic-state cultivation, a technique its researchers proposed in 2024, works on real hardware. On the Willow processor the team measured an upper bound of about 1 × 10-4 on the error of cultivated T-type magic states, roughly 40 times lower than without cultivation, counting only the runs with no detected error; about 8% of attempts were kept. In a separate step it grafted a cultivated state from a small color code into a distance-5 surface code. Magic states supply the non-Clifford operations needed for universal fault-tolerant computation, and they are among its most costly ingredients.
Why the result matters for RSA-2048 and post-quantum planning
Craig Gidney's May 2025 estimate that a machine with fewer than one million noisy qubits could factor a 2048-bit RSA key in under a week relies partly on cultivation to shrink the space spent on magic states. The new paper is the first hardware test of that assumption. It confirms the mechanism at small scale and leaves the scale-up open: the measured error bound of about one in 10,000 sits roughly six orders of magnitude above the one-in-10-billion magic-state quality cited for breaking RSA-2048. This post reads the paper closely and sets it beside NIST's proposed transition dates for post-quantum cryptography and the separate procurement and platform deadlines already running.