IBM claims their quantum computers outperform classical ones

Quantum computing researchers at IBM have announced three demonstrations that they say mark quantum advantage—evidence that quantum computers can outperform classical machines in certain tasks. The world around us is quantum in nature. This makes quantum computers especially promising for modeling it. While a classical computer can approximate how a material would behave or how a certain compound would react with another, experts say that a quantum computer should be able to pinpoint it exactly.

For example, classical computational chemistry methods like density functional theory (DFT) must translate complex quantum behavior into equations that classical hardware can understand. As systems grow more complex, this translation becomes harder, forcing trade-offs in accuracy, speed, or cost. Quantum advantage promises a way to model these processes without compromise.

Quantum computers are expected not only to more accurately model systems than classical computers but also to simulate systems that are beyond the reach of classical computing. “The best examples are simulating quantum materials and breaking certain cryptographic systems,” says Dominik Hangleiter, a computer scientist at the Swiss Federal Institute of Technology (ETH), Zurich, who was not involved in the new IBM demonstrations.

Yet for all their promise, today’s quantum computers remain noisy, are error prone, and easily lose information, and classical computers still outperform them at modeling. Demonstrating quantum computers’ superiority remains a central milestone for the field.

Papers describing all three new demonstrations, which were performed on IBM’s Quantum Heron R3 quantum processors, have been published as preprints, before peer review. On a call with journalists on July 28, researchers from IBM and their collaborators described their results, which show how quantum computers, highly assisted by classical processors, may beat fully classical systems in certain niche problems.

Demo 1: Generating a pattern

Quantum computers use qubits—their fundamental units—which are manipulated by quantum gates (also called operations), just as logic gates control bits in ordinary computers. And the set of available gates defines a quantum computer’s capabilities.

Clifford gates are among the most common gates used in quantum computers. They are easy to simulate but limited in function, and models using only Clifford gates can be efficiently simulated by classical computers.

Adding special gates alongside Clifford gates can make any computation by a quantum computer possible, resulting in what’s known as a universal quantum computer.

Simulating a universal quantum computer on a classical system is believed to be intractable—it would take so long that the result becomes meaningless. Some calculations might require millions of years, or even longer than the age of the universe, even for the best supercomputers. This is where quantum computers are expected to excel.

IBM, in collaboration with University of Chicago researchers, created samples consisting of patterns of zeros and ones using this universal quantum computer. According to the team, no classical computer can replicate the same distribution of patterns within a reasonable time frame (arXiv 2026, DOI: 10.48550/arXiv.2607.25941).

“We mathematically prove that no classical computer can generate this pattern,” IBM scientist Ali Javadi-Abhari tells C&EN.

Hangleiter explains that this mathematical proof shows that at least one set of samples from this quantum algorithm cannot be classically simulated. “It doesn’t mean every one of these samples is hard to simulate,” he says.

The experiment can’t absolutely rule out classical reproduction of the samples that the researchers produced. There might be some unknown shortcut to match the output distribution. Hangleiter says, “All you need is someone with a smart idea.”

But if the sample can’t be reproduced classically, how do researchers know the quantum computer is working correctly?

To verify the results, the IBM and University of Chicago team tested circuits of varying difficulty.

In their model, the researchers added special gates called T gates alongside Clifford gates. But adding T gates physically introduces errors: as more are applied, the computer grows more powerful, but the accumulated noise can render its output meaningless.

With just five T gates, classical simulation was still possible, and the simulations on both quantum and classical computers matched.

Increasing the number of T gates made classical simulation much harder and less accurate. And at the full 468 T gates, no known classical algorithm could simulate the model.

Javadi-Abhari’s group switched approaches: running the quantum computer with no T gates so that a classical machine could predict the output.

Comparing predictions with quantum results, the researchers estimated the circuit’s fidelity at 32%. And then they calculated how much the T gates could lower fidelity. Since the researchers placed them in a way to introduce as little noise as possible, the team found that the fidelity of the quantum computer could drop to, at most, 28.4%.

“If this result were a photograph, fidelity is closer to how closely the entire image matches the original rather than what fraction of pixels are exactly correct,” Javadi-Abhari explains. “Even if the pixels are slightly off color or in the wrong places, they can contribute to a higher fidelity.”

“This is actually a really high number,” Hangleiter says. It is 100 times as good as prior leading work. He thinks this demonstration is unique, and he would like to see more similar research.

Demo 2: Kicking a quantum magnet

Researchers at Qedma, a quantum computing start-up, harnessed an IBM quantum computer to simulate a quantum magnet—a material whose magnetic behavior is governed by quantum mechanics in ways classical physics can’t explain (arXiv 2026, DOI: 10.48550/arXiv.2607.24937).

Each atom has its own magnetic moment—imagine a tiny compass needle. In a regular magnet, magnetism arises when many atoms align these moments in the same direction.

In a quantum magnet, an atom’s magnetic moment can point in a mix of directions at once. This is called quantum superposition—the same strange idea behind Schrödinger’s cat being both alive and dead at the same time. The magnetic moments of neighboring atoms can also become entangled, linking their behavior.

All these entangled, superposed—quantumed-up—magnetic moments can act together in strange ways, producing exotic states of matter. Cracking this behavior could help explain high-temperature superconductivity, potentially transforming energy transmission and electronics.

At Qedma, Netanel Lindner’s team simulated a quantum magnet being kicked at regular intervals.

A kick is a sudden jolt to the magnet’s state—like hitting it with a pulse of light. On the quantum computer, each kick was created by applying a series of gates to all the qubits.

When a system is kicked repeatedly, it can’t immediately settle into thermal equilibrium. Instead, it lingers in an in-between phase called a prethermal state, where it stays structured before finally relaxing to equilibrium. Think of shaking a snow globe: for a while, the snow is suspended in a “prethermal state” before drifting back down.

This prethermal window is when the magnet’s most intriguing quantum effects appear. Magnetization during this stretch is especially useful for understanding its behavior. Those same effects also make the magnet difficult for classical computers to simulate accurately.

Lindner’s team simulated different sizes of quantum magnets being kicked and compared quantum results with those of classical simulations for the same sizes.

For 28-atom magnets, the classical computer’s results matched both physical intuition and the quantum computer’s—showing that the quantum computer works as expected.

But for 51- and 74-atom magnets, three classical models failed: they ran out of memory, lost accuracy as things got complicated, or became too slow to be practical. The quantum processor, however, kept going—showing magnetization values that matched theoretical predictions.

This limitation of classical computers makes it hard to verify the quantum results for larger magnets—no one knows the exact magnetization values to check against. The researchers can’t compare the simulation with a real material because the magnets were a theoretical system tailored for the IBM processor. “It’s like choosing a problem that is really easy to simulate on a quantum computer instead of something you would find in nature,” Hangleiter says.

However, Lindner’s group ran its simulation on several quantum processors, each with its own noise patterns and error mitigation systems, including one that ran on different hardware, to check for errors. The results matched.

Getting the same magnetization behavior on such different machines makes it less likely the results are due to a hardware-specific glitch.

Demo 3: Measuring a realistic system

In the third demonstration, researchers at quantum software company Algorithmiq simulated a material on an IBM quantum computer. A material that, like Qedma’s magnet, exists only in theory (arXiv 2026, DOI: 10.48550/arXiv.2607.25998).

To make it behave more like real matter, Sergei Filippov and his team deliberately introduced disorder and studied how information flowed through the system.

“Catalysts, batteries, and all materials that will power the next generation of clean energy are messy, disordered, and irregular, and this messiness is what makes them hard to simulate,” Sabrina Maniscalco, cofounder and CEO of Algorithmiq, said during the press briefing. “We designed a model of real-world disorder in matter.”

Just as you can infer how transparent or opaque a material is by shining light on one end and seeing how it comes through the other end, Filippov’s team disturbed one group of qubits—each representing an atom—and measured the effect on a group of qubits at the other end of the computer after the disturbance traveled through it.

Because of the disorder in the system, information spread unevenly: some regions let it pass quickly, while others slowed it down.

In classical physics, you can measure a system before and after a disturbance or process to see what changed. But in quantum mechanics, measurement itself alters the system. Checking a property like the position of an electron or a qubit’s state destroys the quantum state you need, making it impossible to measure the same system twice and directly compare results.

Instead, the researchers used a clever workaround to measure how information flow changed the material: reversing the simulation. First, they applied a sequence of operations to simulate the material’s natural evolution in time. Then, they introduced a small disturbance and finally ran all the operations backward, reversing the flow of time in the material.

Without a disturbance, reversing the simulation returns the qubits to their initial state. But the disturbance disrupts this perfect rewind. As the system ran backward, only the effects linked to the disturbance remained.

Measuring these changes produces a value called the operator Loschmidt echo. Much as echoes in a cave can give information about the shape of the cave, this Loschmidt echo reveals key information about the material.

Filippov’s team ran the simulations on two different IBM quantum computers and also used two classical algorithms on a classical processor.

Across different levels of disorder, the two quantum computers produced similar echo signals, while the classical simulations disagreed with each other and with the quantum results. This suggests that the quantum simulations stayed accurate where classical methods faltered. Agreement between the quantum processors, each with different noise and error profiles, also makes it less likely the signal was caused by one machine’s quirks.

However, both quantum processors share the same architecture, and there’s no external benchmark to check the results. The material was tailored for the IBM processor and doesn’t correspond to any real substance researchers can make in the laboratory.

So verification remains internal: the IBM quantum computers check each other, while classical methods disagree. Because the simulated material was built for the hardware, it’s hard to say if matching results reflect real physics or just suit the machines they were designed for.

“I don’t know the details enough to know if I trust it,” Hangleiter says. The researchers verify the outcome by characterizing the noise in the processor really well, he says. So they have good control over what’s happening, and they leverage that to have an idea of what the actual error of your computation would be. “That may be a reason to trust it.”

Is this quantum advantage? What does that mean for chemistry?

Whether these demonstrations achieved quantum advantage is unclear, according to Hangleiter. “It is unlikely that we will ever definitively prove quantum advantage,” he says.

In all three demonstrations, a repeating concept is that it is difficult to verify the correctness of a quantum computer’s results. “The verification is better than what we’ve seen in other demonstrations so far, and they’re really making a good effort,” Hangleiter says. But there’s no way to 100% prove the researchers are correct.

One way to verify a quantum computer’s results would be to simulate a real experiment and directly compare the results with nature.

“This is our and everyone’s goal here,” said Director of Research and IBM fellow Jay Gambetta at the press briefing this week. Gambetta also said that the goal in publishing these three papers was to build trust in quantum computing.

So even though these demonstrations are extremely niche applications that are not commercially usable in any way, Hangleiter says he thinks that these new results build that trust and sharpen the case for what quantum computing can do.

Lindner says he views the new demonstrations as “a shift towards practical quantum computing that people can use today for scientific applications.”

Lindner says that theoretical physicists today should be using quantum computers. In time, he expects this will hold true for chemistry and other scientific fields. “Chemistry is one of the most important long-term applications of quantum computing,” he says.

“But the path to practical advantage for chemistry depends strongly on further advancements in quantum hardware,” Lindner says. Meanwhile, “I would encourage chemists to engage with quantum computing by strengthening the community’s understanding of what questions can be efficiently addressed by quantum computers.” Classical computational chemistry isn’t being replaced. “This result is proof of principle for an additional scientific instrument.”