Google’s Willow Wins a Quantum-Only Game 98.3% of the Time

Google’s Willow quantum processor won the “magic square game” 98.30% of the time, against a classical ceiling of 88.89%, Google Quantum AI said on 19 November.

The results are published in Science. Google says its qubits also beat classical strategies in tasks spanning up to 105 qubits.

Business Pill · Vertical Integration

An evergreen Business Engineer explainer on vertical integration. It teaches the general idea only and does not discuss Google’s quantum work.

The key insight: Google turned a quantum effect into a score: Willow won 98.30% of rounds where classical play is capped at 88.89%, and stayed above classical bounds up to 105 qubits.

What Google Measured

The game asks two players to fill a 3×3 grid of +1 and −1 under rules no fixed table can satisfy. Any pre-written plan wins at most 8 rounds in 9, Google says.

Players sharing entangled qubits can in principle win every round. Google played all nine versions 100,000 times each on Willow; noise caused about one loss in 60.

On our arithmetic, that is about 1.7% of rounds lost against at least 11.1% for the best classical strategy, roughly 6.5 times fewer losses.

Willow won the magic square game 98.30% of the time; no classical strategy can beat 88.89%, per Google Quantum
Willow won the magic square game 98.30% of the time; no classical strategy can beat 88.89%, per Google Quantum AI.
Google diagram of the quantum magic square game with the 98.3% hardware win rate against the 88.9% classical bound
The magic square game on Google’s hardware: measured win rate against the classical bound. Image: Google Quantum AI, Google Research blog, 19 Nov 2026.

A Number for Contextuality

The property behind the win is called contextuality: a measurement’s result can depend on which other compatible measurements are made with it. Google calls it “arguably the strangest feature of quantum mechanics”.

Google bundled the correlations into one score, χ. Predetermined outcomes cap it at 4; quantum mechanics allows up to 6. Google measured χ = 5.590 ± 0.008, about 7% below the quantum limit, the post says.

Does It Survive at Scale?

In a multiplayer parity game, Google ran up to 71 players, one per qubit, and stayed above the classical bound throughout. It also verified genuine entanglement across 45 qubits.

Google also ran a 2018 hidden-linear-function algorithm on grids of up to 105 qubits. Success probability fell from about 1 to about 0.6 at full size, still above the conjectured classical bound, per the post.

Google plot of GHZ parity game win probability from 3 to 71 qubits above the classical bound
The multiplayer parity game from 3 to 71 qubits, above the classical bound throughout, per Google. Image: Google Quantum AI, Google Research blog, 19 Nov 2026.

Why Google Cares

Google says the tasks became “a sensitive new benchmark” for quantum hardware, catching subtle errors that random-circuit tests miss. It links that to applications such as chemistry and materials simulation.

The post also ties contextuality to the “magic states” error-corrected machines need, which Google says its team recently prepared (our October piece).

The Structural Read

As we read it, the business value Google claims here is measurement: a benchmark with a fixed physical ceiling that exposes errors other tests average away.

As we read the magic-state link, Google is connecting a physics result to its error-correction roadmap rather than to a product.

Google Quantum AI, 19 Nov 2026

“a sensitive new benchmark for quantum hardware”

Three Implications

FOR QUANTUM HARDWARE MAKERS A benchmark with a ceiling set by physics gives a fixed target, per Google.

FOR GOOGLE The work links contextuality to the magic states its error-corrected roadmap depends on.

FOR BUYERS OF QUANTUM COMPUTE Google names chemistry and materials simulation as where hidden errors matter most.

What Is Not Established

Google gives no commercial timeline, no customer and no product in the post. The 105-qubit result is measured against a conjectured classical bound, not a proven one.

This Week In The AI Supercycle

AI Lost Its Central Banker, the Lab Oligopoly, and the Future of Open Source

This week’s issue: the Nvidia leadership change, read as the AI build-out losing its central banker: who now finances the GPU landlords, backstops the data centres and sponsors open models, and why two labs are becoming the bank.

Read the full issue →

Business Engineer Framework

The AI Supercycle

The Business Engineer map of where value pools in the AI stack.

Read the Map of AI →

The Bottom Line

Google’s new paper puts numbers on a quantum effect at up to 105 qubits: a 98.30% win rate against an 88.89% classical ceiling, and a new yardstick for its own hardware.

97,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.

A note on sourcing. Every fact in this piece is from Google Quantum AI’s 19 November blog post describing its paper in Science. The 1.7%, 11.1% and 6.5x figures are our arithmetic on Google’s numbers. We have not seen independent replication.

Sources: Google Research blog, ‘A game you cannot win unless you have a quantum computer’, 19 Nov 2026 · Science, ‘Quantum-classical separation in bounded-resource tasks arising from measurement contextuality’

Scroll to Top
FourWeekMBA