Google Willow Quantum Processor Clears Classical Ceiling in Magic Square Game

Google’s Willow quantum processor won a four-qubit coordination game 98.3% of the time, clearing a classical ceiling of 88.9%, researchers reported Thursday in the journal Science.

The Mermin-Peres magic square game assigns two players intersecting parts of a three-by-three grid without allowing them to communicate. One player completes a row while the other completes a column, each using +1 or -1, and their entries must meet contradictory parity conditions while agreeing on the shared square, BigGo Finance reported. No pre-arranged table of answers can cover all nine cases.

Willow’s edge comes from quantum contextuality, a property under which measurement outcomes depend on which other measurements are performed alongside them. "You cannot assign fixed, pre-existing values to everything you might measure," the researchers wrote. “Our qubits coordinated better than any pre-written table allows.” Unavoidable hardware noise kept the run short of the theoretical ideal.

The team also put a number on contextuality itself. Any model assuming predetermined outcomes caps out at 4, while quantum theory permits 6. Willow registered 5.590, with a stated uncertainty of 0.008.

Larger tests, narrower margins

In a separate run, the processor executed a GHZ parity game that asked up to 71 entangled qubits to produce coordinated binary answers. Success was measured against the best classical winning rate, which approaches 50% as the player count increases, and the hardware remained above that threshold the entire time. The researchers confirmed genuine entanglement across 45 qubits.

A third task, the hidden linear function problem, required a shallow quantum circuit to output a binary string satisfying mathematical constraints. With 105 qubits, the processor achieved roughly 60% success, versus an expected ceiling of about 50% for a classical approach of comparable depth.

Task Qubits used Google result Classical limit
Magic square game 4 98.3% win rate 88.9%
GHZ parity game Up to 71 Above classical ceiling throughout Approaches 50%
Hidden linear function 105 ~60% success ~50%
Contextuality score — 5.590 ± 0.008 4

Diagnostic tool, not a commercial milestone

Google is careful to frame the work as a diagnostic. The company’s corporate blog described contextuality testing as "a sensitive new method to catch subtle hardware errors," potentially giving engineers a sharper tool for evaluating the circuits behind chemistry and materials research. The experiments exposed small systematic errors that conventional randomized benchmarks can mask.

href="http://www.youtube.com/watch?v=kQJo7Wd-_8w&t=6s" class="lQkvvf">www.youtube.com/watch?v=kQJo7Wd-_8w&t=6s Posted on X 20/11/26 3:51 PM

The result does not mean Google holds a commercially useful quantum computer. The headline 98.3% came from just four qubits. The 105-qubit run tested a different and more limited challenge, and its largest instance still failed in roughly four out of ten attempts.

For crypto markets, the long-run concern is that a future fault-tolerant quantum computer could break the digital signatures securing Bitcoin wallets and transactions. Nothing in the new findings changes that timeline. The 105-qubit demonstration was a shallow mathematical circuit, not an attack on cryptography.

The findings appeared Thursday in Science, the peer-reviewed journal of the American Association for the Advancement of Science that has been published weekly since 1880 and carries a 2024 impact factor of 45.8. With an estimated readership exceeding 400,000 and an acceptance rate below 7%, placement in its pages signals that a result has survived intense scrutiny across all scientific disciplines.

The more immediate significance is methodological. Games with provable classical ceilings offer a cleaner yardstick, and the contextuality measurement gives engineers a signal that can flag drift before it corrupts larger computations. Whether that yardstick scales is the open question.

The four-qubit game is small enough to verify exhaustively, while the larger tasks traded precision for size as noise accumulated. Google’s results show the processor clearing classical bounds at each scale it tested, but the margins narrow as the circuits grow.