Skip to content
Quantum.world

The Signal

Google says contextuality-linked task ran on up to 105 qubits

Google Quantum AI researchers report measuring quantum contextuality on a superconducting processor and running a related hidden-linear-function task on grids of up to 105 qubits. They report a 98.30% win rate in a two-party game, above its 88.89% classical limit.

Google says contextuality-linked task ran on up to 105 qubits

Google Quantum AI researchers report measuring quantum contextuality on a superconducting processor and running a contextuality-linked task on grids of up to 105 qubits. The findings appear in a paper published in Science, “Quantum-classical separation in bounded-resource tasks arising from measurement contextuality.”

Testing the classical limit

The team used Google’s Willow processor for a version of the magic square game. In each round, two players receive a row and a column of a three-by-three grid and must give matching answers at their intersection, while following different rules for the products of their answers.

A strategy fixed in advance cannot satisfy every rule: the best classical approach wins at most 88.89% of rounds. Google’s researchers report that the processor won 98.30% of the time, with a stated uncertainty of 0.0001. They played all nine versions of the game 100,000 times each; the post describes the hardware as losing roughly one round in 60.

A direct contextuality measurement

The researchers also measured contextuality using a Bell–Kochen–Specker inequality, which combines correlations into a value called chi. Outcomes fixed in advance must obey a classical bound of chi no greater than 4; quantum mechanics permits a value as high as 6.

On the processor, the team reports a value of 5.590, with an uncertainty of 0.008. The measurement used two main qubits and an extra helper qubit, or ancilla, to record results while leaving the main qubits intact. The team ran 540 measurements in sequence and repeated that sequence 1,000 times.

Scaling to larger systems

For a separate GHZ parity game, the researchers used systems with up to 71 players, one for each qubit. They report that the measured win rate remained above the best classical strategy as the number of players grew, though noise reduced the rate. The team also reports verifying genuine entanglement across 45 qubits.

The largest task tested was a hidden-linear-function problem, which the researchers ran on grids of up to 105 qubits. They report success probabilities above one-half, which the post places above its conjectured classical bound. At the largest processor size, the reported probability fell from about 1 to 0.6 as the system grew.

A benchmark for hardware

Google’s researchers say the tests exposed subtle coherent errors that random-circuit benchmarks can average away. They propose using the tasks as a hardware benchmark: the magic-square game has a maximum win probability of 1, while the contextuality measure has a quantum limit of 6.

The post, dated November 19, 2026, is by Shashwat Kumar and Rodrigo Cortiñas, research scientists at Google Quantum AI. It says the work was conducted by the Google Quantum AI team with academic collaborators.

Sources

  1. Original source · Blog post Google Research research.google

Related stories

Explore

More articles