Google’s Willow quantum processor wins 98.3% in four‑qubit magic‑square test
Google reported that its four‑qubit processor Willow achieved a 98.3% success rate in the magic‑square game, surpassing the 88.9% classical limit.
Key takeaways
- Willow won the four‑qubit magic‑square game 98.3% of the time, above the 88.9% classical ceiling.
- Google also ran a 71‑qubit GHZ parity game and a 105‑qubit shallow‑circuit task, achieving roughly 60% success on the latter.
- The experiment measured quantum contextuality, recording a value of 5.590 ± 0.008, exceeding the classical bound of 4.
Google announced that its four‑qubit quantum processor, Willow, succeeded in the Mermin‑Peres magic‑square game 98.3% of the time, exceeding the 88.9% maximum success rate achievable by any classical strategy, according to a Science paper released on Thursday, 22 November 2026.
Experimental details and scale
The magic‑square test involved 900,000 individual trials, each requiring two players to fill intersecting parts of a three‑by‑three grid without communication while obeying conflicting row‑and‑column constraints. Classical pre‑arranged tables can win at most eight of nine assignments, which translates to an 88.9% ceiling; entangled qubits can in principle win every round.
Willow’s hardware produced a 98.3% win rate, with the shortfall attributed to unavoidable hardware noise that prevented reaching the quantum ideal of 100%. The same study reported that Willow measured contextuality at 5.590 ± 0.008, where the classical limit is 4 and the theoretical quantum maximum is 6.
In a separate GHZ parity game, Google linked up to 71 entangled qubits to deliver coordinated binary answers. The strongest classical winning probability for that coordination game approaches 50% as the number of participants grows, yet Willow’s system stayed consistently above that threshold throughout the run.
A third benchmark, described as a hidden linear function task, used a shallow quantum circuit of 105 qubits to generate binary strings that satisfy specific mathematical constraints. The processor succeeded in roughly 60% of the attempts, meaning it failed about four out of ten runs.
Contextuality testing and prior progress
Google researchers emphasized that the contextuality measurement reveals correlations that cannot be explained by assigning fixed, pre‑existing values to all observables, a hallmark of quantum behavior. The paper states, “our qubits coordinated better than any pre‑written table allows,” underscoring the advantage of entanglement over classical strategies.
The authors also noted genuine entanglement across 45 qubits during the experiments, confirming that large‑scale coherence was maintained despite increasing noise with system size.
In October 2025, Google reported a 40‑fold reduction in error rates when preparing magic states, a resource required for fault‑tolerant quantum computing. The current results build on that improvement, using contextuality testing as a “sensitive new method to catch subtle hardware errors,” according to Google’s corporate blog.
While the 105‑qubit shallow‑circuit benchmark does not constitute fault‑tolerant computing, the researchers argue that the tests expose systematic errors that conventional randomized benchmarking can miss, potentially aiding future applications in chemistry and materials science.


Discussion 0