Technician seen from behind at a wafer inspection microscope in a semiconductor cleanroom, no faces, logos or readable text
AI-generated image

Research

Google’s quantum chip just won an “impossible” game 98.3% of the time, and no classical strategy can beat 88.9%

Published · Last updated

Entanglement gets the headlines, but Google thinks something stranger is doing the work. "Entanglement usually steals the show in quantum computing, but theory points to a deeper resource: quantum contextuality," Google Quantum AI posted on X on Thursday. "In Science, we run contextuality-based tasks up to 105 qubits on Willow, creating a fixed, physics-based benchmark for quantum hardware."

## The game

The headline result comes from the magic square game. As Google's research blog explains it, two players, Alice and Bob, each fill in part of a 3×3 grid of +1s and −1s under rules that no real grid can satisfy at once. Without talking, they win a round if their answers match where the row and column cross. Any plan written in advance wins at most 8 rounds in 9, or 88.89%.

With shared entangled qubits, the odds change. Google says it played all nine versions of the game 100,000 times each on its Willow processor and won 98.30% of rounds. Sundar Pichai summed up the results on X: the "impossible" game won 98.3% of the time, a score of 5.59 on a contextuality test where the classical bound is 4 and the quantum maximum 6, wins over classical strategies in multiplayer games of up to 71 qubits, and a "landmark shallow-circuit algorithm" run at scale for the first time on up to 105 qubits.

## What contextuality means

Contextuality is the idea that a measurement's result is not fixed in advance but can depend on which other compatible measurements are made alongside it. Theory suggests it may be what powers some quantum algorithms. Until recently it had mostly been shown in systems of a few particles. The paper in Science tests whether it survives on a full processor.

## A ruler, not just a record

Google's most practical claim is about testing hardware. These tasks need many qubits to agree exactly, and Google says running them exposed small coherent errors that standard random-circuit benchmarks tend to average away. Because the ceiling is set by physics (a score of 6, or a win rate of 100%), the gap between a machine's score and that ceiling gives what Google calls an objective benchmark.

That fits Google's recent run of Willow papers, including its [magic state cultivation result](/news/google-magic-state-cultivation-willow-science.html). This one is not a claim of useful quantum advantage, and Google does not present it as one.

← All news