Google’s Quantum AI team has successfully demonstrated the “magic state cultivation” technique in practice on the 105-qubit “Willow” processor, reducing error rates by roughly 40-fold and achieving a record logical accuracy of 99.99%. Published in Science on October 15, 2026, this milestone marks the first practical solution to one of the biggest computational bottlenecks in quantum computing.
Scientific breakthrough on the Willow processor cuts errors forty-fold
Mechanisms of magic state cultivation and transitioning between error-correction codes
Widespread academic praise and the challenges of cracking encryption and simulating molecules
A foundational step toward fault-tolerant quantum computers
Frequently asked questions
Scientific breakthrough on the Willow processor cuts errors forty-fold
Google’s Quantum AI division has proven that the “magic state cultivation” technique, which the company’s researchers first proposed theoretically in 2024, works fully in practice on real hardware. On the 105-qubit “Willow” processor, the technique reduced error rates by about 40-fold compared to uncultivated states, according to a research paper published on October 15, 2026, in the prestigious scientific journal Science.
The research team announced that it successfully produced “T-states” with a record logical accuracy of 99.99%. In an official announcement by Google, Alphabet’s quantum computing unit described the work as “the first practical demonstration of an application solution” for one of the worst resource bottlenecks in the field of quantum computing.
Magic states are a necessary element for performing computations that give a quantum computer the ability to execute tasks beyond the reach of traditional classical machines. However, producing them is extremely costly; estimates suggest their generation can consume up to 95% of a quantum computer’s total capacity, noting that each state can only be used once.
Mechanisms of magic state cultivation and transitioning between error-correction codes
The currently prevailing standard method, known as “distillation,” relies on combining multiple imperfect states to produce a smaller number of higher-accuracy states. In contrast, the “cultivation” technique works on a single, isolated state; the team encoded an initial crude magic state within a few qubits, checked whether it was the correct state, and then deployed the states that passed the check across a larger number of qubits to provide stronger protection. The experiment used 54 of the 105 physical qubits featured in the Willow processor. Without cultivation, roughly 4 out of every 1,000 attempts produced a defective state, according to Scientific American.
The researchers also successfully transferred states from a distance-3 “color code” to a distance-5 “surface code,” two different systems for quantum error correction. Google stated that this proves the purified states can survive the transfer process on its hardware.
During testing, lead researcher Emma Rosenfeld discovered that laboratory noise caused by equipment drift and temperature changes was distorting the results of a standard measurement metric. To overcome this, the team devised a clever method using the cultivation technique to self-measure.
Widespread academic praise and the challenges of cracking encryption and simulating molecules
The innovation received notable scientific praise. Isaac Kim of the University of California, Davis, told New Scientist that this approach “could make fault-tolerant quantum computing much cheaper.” Dominic Hanglieter from ETH Zurich described it as “a potential game-changer” for machines comprising hundreds of logical qubits.
Nonetheless, this method only retains states that pass strict quality checks; only 8% of run rounds produced usable magic states. Jinyu Cai of Imperial College London noted that error rates on the Willow processor must drop further before it can handle practically useful problems.
The team currently stands at an error rate of roughly 1 in every 10,000 attempts, while breaking RSA-2048 encryption requires a failure rate close to 1 error in every 10 billion attempts, and simulating chemical molecules requires an error rate between 1 in a million and 1 in 100 million attempts, according to Scientific American. Google acknowledged the need to prove the technique on a larger scale, with Rosenfeld stating: “This is the first demonstration proving the concepts work; but we are not done yet.”
A foundational step toward fault-tolerant quantum computers
The success of the magic state cultivation technique on the 105-qubit Willow processor, alongside the successful transfer of purified states between color and surface error-correction codes, provides tangible evidence of the possibility of overcoming one of the most complex computational bottlenecks threatening to consume 95% of quantum machine capacity.
Although the road still requires reducing error rates from their current level of 1 error in every 10,000 attempts to much finer levels to achieve complex chemical molecule simulation or crack RSA-2048 encryption, this innovation opens a less costly and more efficient path for the next generation of quantum computers.
Frequently asked questions
Question: What achievement did Google’s Willow chip make in the study published in Science?
Answer: It successfully applied the magic state cultivation technique in practice on 54 out of 105 qubits in the Willow processor, reducing errors 40-fold and achieving a logical accuracy of 99.99%.
Question: What is the difference between the new cultivation technique and the traditional magic state distillation method?
Answer: The distillation method combines multiple flawed states to produce a fewer number of precise states, whereas cultivation inspects a single state and expands its protection across a larger number of qubits to conserve resources.
Question: What is the current percentage of successful attempts and the recorded error rate in the experiment?
Answer: Roughly 8% of run rounds produced usable magic states with an error rate of 1 in every 10,000 attempts, compared to 4 errors in every 1,000 attempts without cultivation.
Question: What future accuracy rate is required to simulate molecules or crack digital encryption?
Answer: Simulating molecules requires an error rate between 1 in a million and 1 in 100 million, while breaking RSA-2048 encryption requires just 1 error in every 10 billion attempts.
