Google Willow

Google Willow is a superconducting circuits quantum computer operated by Alphabet in USA. It has 105 physical qubits. It ranks 20 of 43 superconducting machines in this table by physical qubit count. 11 sourced specs are in the spec sheet. 6 published benchmark or milestone records are listed below, each marked by who measured it.

Deep dive: Google Willow: what it has and has not shown

Also known as Willow chip

Google Quantum AI's 105-qubit chip, announced on 9 December 2024. Google reports T1 times approaching 100 microseconds and a surface-code memory scaled from 3x3 to 7x7 that halved the logical error rate at each step. Access outside Google has been very limited; Google opened an early-access proposal round with a 15 May 2026 deadline. The error correction result covers a memory, not logical gate operations.

Spec sheet

Technical details as the builder or a paper states them, each with its own source. A blank is a fact we could not source, not a zero. Fidelity and coherence figures are in the benchmark table below.

Spec	Value	Basis	Note	Source
Processor	Willow, 105-qubit superconducting chip	Vendor reported	Spec sheet published 9 Dec 2024 lists two Willow chips, one tuned for error correction and one for random circuit sampling.	quantumai.google blog.google
Qubit type	superconducting transmon	Peer reviewed	Both are Google papers; Nature text calls it a square grid of superconducting transmon qubits.	arxiv.org nature.com
Topology	square grid, average connectivity 3.47 (4-way typical)	Vendor reported	Spec sheet gives 3.47 as the average number of neighbours per qubit on the 105-qubit device.	quantumai.google arxiv.org
Max connectivity	4	Vendor reported	Spec sheet lists 4-way connectivity as typical; edge qubits have fewer neighbours.	quantumai.google
Native gates	microwave XY and virtual Z single-qubit gates, CZ entangling gate	Peer reviewed	Applies to the error-correction chip; the spec sheet says the random-circuit-sampling chip uses an iSWAP-like two-qubit gate.	arxiv.org
Single-qubit gate time	25 ns	Peer reviewed	arXiv supplement: all XY rotations last 25 ns on the 105-qubit processor.	arxiv.org
Two-qubit gate time	42 ns	Peer reviewed	arXiv supplement: CZ gates take 42 ns on the 105-qubit processor, 37 ns on the 72-qubit processor.	arxiv.org
Fabrication	Google's Santa Barbara quantum chip fabrication facility	Vendor reported	Google says Willow was fabricated in its new facility in Santa Barbara.	blog.google
Access and pricing	early access by research proposal reviewed by Google Quantum AI; no price stated	Independent	QCR reports a 15 May 2026 proposal deadline and 1 July 2026 notifications; no pricing is mentioned.	quantumcomputingreport.com
Predecessor	Google Sycamore	Vendor reported	Blog calls T1 about 5x better than the previous generation; Nature paper compares to earlier Sycamore processors.	blog.google nature.com
Roadmap	next goal is a useful beyond-classical computation, then a large error-corrected quantum computer	Vendor reported	Google's December 2024 post places Willow on a six-milestone roadmap ending at a large error-corrected machine.	blog.google

Published benchmark figures

These are numbers the builder or a paper reports for this machine, not results from a common test. The basis column says who measured them. What the metrics mean and why they are not comparable across technologies.

Metric	Value	Measured	Basis	Note	Source
Energy relaxation time (T1)	100 µs	2024-12	Vendor reported	Google says T1 times are approaching 100 us, about 5x the previous generation; approximate figure	blog.google

Milestone claims

Nature: RL control of error correction on Willow; 3.5x better surface-code stability under injected drift; logical error per cycle 7.72e-4 (surface) and 8.19e-3 (colour) (2026-07-08, peer reviewed) nature.com
Google claimed verifiable quantum advantage: Quantum Echoes (OTOC) ran 13,000x faster on Willow than the best classical algorithm on a top supercomputer (2025-10, vendor reported) blog.google
Nature: colour code on 72-qubit Willow, distance 3 to 5 suppresses logical error 1.56x (below colour-code threshold per simulation); magic-state fidelity over 99% with post-selection (2025-05-26, peer reviewed) nature.com research.google
Google claimed a random circuit sampling task in under five minutes that it estimates would take a leading supercomputer 10^25 years (2024-12, vendor reported) blog.google
Below-threshold surface code memory, distance 7: 0.143% logical error per cycle, error suppression factor 2.14 per distance step (Nature 2025) (2024-08, peer reviewed) arxiv.org blog.google

Timeline

2024-12-09: Announced
2026-10: Status checked: operational

Comparable superconducting machines

Closest in physical qubit count within the same technology. Counts of different qubit types are not interchangeable across technologies.

Machine	Operator	Qubits	Status
Zuchongzhi 3.0	USTC	105	operational
Zuchongzhi 3.2	USTC	107	operational
Rigetti Cepheus-1-108Q	Rigetti Computing	108	operational
IBM Nighthawk	IBM	120	operational
IBM Eagle	IBM	127	decommissioned
University of Tokyo IBM Quantum System One	University of Tokyo	127	operational