What happened?
On October 15, 2026, Google Quantum AI reported in Science that its Willow quantum chip produced magic states, a special ingredient quantum computers need, with a record logical fidelity of 99.99% Google blog ↗ Science ↗. Google says the method cut errors about 40 times compared with states that were not cultivated Google blog ↗.
Quantum computers can already run some operations, called Clifford gates, with strong error protection. But those gates alone cannot run every algorithm. To unlock the rest, a machine needs extra resources called magic states, also known as T states Google blog ↗ arXiv ↗.
Magic states are expensive. Google says up to 95% of a future quantum computer's capacity could go to producing them, and each one can be used only once before a new one is needed Google blog ↗. Making them cheaper and cleaner is one of the biggest open problems on the road to useful quantum computers.
In 2024, Google researchers Craig Gidney, Noah Shutty and Cody Jones proposed a cheaper method called magic state cultivation arXiv ↗. Instead of the older approach, called distillation, which combines many noisy states into one good one, cultivation grows a single state inside a protected patch of qubits and checks it step by step, throwing it away if errors show up arXiv ↗. Their paper argued this could cost roughly as much as an ordinary two qubit operation arXiv ↗.
The new Science paper is the first time Google has shown this idea working on real hardware at this quality Google blog ↗ Science ↗. The team also showed it could move, or "graft", a cultivated state from one type of error correcting code into another, larger one that the rest of the computer would use Google blog ↗.
How does it work?
A logical qubit is built from many physical qubits working together so that errors can be spotted and fixed. Cultivation starts a magic state inside a small logical qubit, then repeatedly measures checks on it. If any check signals an error, the attempt is thrown out and restarted. States that pass every check are far cleaner than the raw hardware alone could make Google blog ↗ arXiv ↗.
All three figures are Google's own reporting Google blog ↗.
Google says the team grafted the state from a distance 3 color code to a distance 5 surface code Google blog ↗. In plain terms, the state was made in a small, convenient code and then handed over to the larger, sturdier code that a full machine would use for computing. That handover is important, because a magic state is only useful if it can join the rest of the computation Google blog ↗.
Getting there took careful engineering. Emma Rosenfeld, one of the authors of Google's post, found that measurement problems were caused by drift in the equipment and by temperature changes Google blog ↗ Interesting Engineering ↗. The team fixed them with better control of the hardware and a new way to benchmark the results Google blog ↗. This was a physics and engineering result; Google did not describe an AI system as part of the method.
Which fields could this affect?
This is a step in building quantum computers, not a product. These connections are our assessment.
Quantum computing research
Labs racing to build error corrected machines now have a demonstrated, cheaper path to magic states on real hardware Google blog ↗ Science ↗.
Explore scienceChemistry and new materials
Simulating molecules is a leading hoped for use of quantum computers, and such algorithms need many magic states Google blog ↗. Cheaper states would make those runs more realistic.
Encryption and security
Breaking today's public key encryption would also need huge numbers of magic states. This result does not bring that close; it only lowers one cost on a long list Google blog ↗ Interesting Engineering ↗.
What has been checked?
The evidence is a peer reviewed Science paper, Google's own explanation, the 2024 theory paper and news coverage. Leapscope reviewed these sources; we did not repeat or check any measurements.
Shown so far
- The result was published in Science on October 15, 2026 Science ↗.
- Google reported a logical magic state fidelity of 99.99%, which it calls a record Google blog ↗.
- Google reported about 40 times fewer errors than states that were not cultivated Google blog ↗ Interesting Engineering ↗.
- The team moved a cultivated state from a color code into a larger surface code Google blog ↗.
Still unknown
- Whether the method holds up at much larger scale; Google says that still has to be shown Google blog ↗.
- How often attempts are thrown away and restarted on real hardware, and what that costs in time.
- When a full fault tolerant quantum computer will exist; Google says plenty of technical challenges remain Google blog ↗ Interesting Engineering ↗.
Evidence status: Peer reviewed paper. Stage: Verified. Peer reviewed and published in Science; still needs to be shown at larger scale.
From one clean state to a working machine
This is our suggested way to follow the story, not a promised timetable.
Can I use it today?
You can read Google's explanation, the Science paper record and the 2024 theory paper Google blog ↗ Science ↗ arXiv ↗. The Willow chip is not something the public can use for everyday tasks.
A few things you might be wondering
What is a magic state in quantum computing?
A magic state is a special prepared quantum state that lets a quantum computer run the operations it cannot do with its basic protected gates. Without them, a quantum computer cannot run every algorithm Google blog ↗ arXiv ↗.
What is magic state cultivation?
It is a method proposed by Google researchers in 2024 that grows one magic state inside a protected patch of qubits and checks it repeatedly, discarding it if errors appear arXiv ↗. It aims to be much cheaper than older distillation methods arXiv ↗.
Did Google build a fault tolerant quantum computer?
No. Google says plenty of technical challenges remain and the method still needs to be shown at larger scale Google blog ↗.
How good were the magic states?
Google reported a logical fidelity of 99.99% and about 40 times fewer errors than states that were not cultivated Google blog ↗.
Is AI involved in this result?
Google did not describe an AI system as part of this method. It is a quantum hardware and error correction result Google blog ↗ Science ↗.
Go straight to the sources
Checked Oct 8, 2026. The first source is the original announcement or research. Later sources add independent context; background pages do not validate the result on their own.
01Google's explanation of the 99.99% fidelity result, the 40 times error reduction and the grafting step.
The peer reviewed paper describing the experiment on Google's Willow processor.
News coverage of the result, including the measurement problems the team had to solve.
The original theory paper by Craig Gidney, Noah Shutty and Cody Jones.