IBM says 2026 was the year quantum computers beat regular ones

A quantum computer uses qubits, the quantum version of the bits in your phone. Qubits can hold a blend of 0 and 1 at the same time. That lets a quantum computer tackle some problems in a very different way.

Quantum advantage is the moment one of these machines does a task that classical computers, meaning the regular kind, cannot match. IBM predicted two years ago that this would happen by the end of 2026. It now says it did.

• The catch: advantage is a milestone, not a finish line.
• IBM's own words: the race is now to make quantum computers useful.

Why it matters: a one-off win is a science result. Useful work in chemistry, materials and finance needs far more than a single benchmark.

Nighthawk r2 is IBM's newest processor. A processor is the chip that does the thinking. This one has 120 programmable qubits, which sit on a chip with 458 physical qubits in total. They are laid out in a square grid, so most qubits link directly to four neighbors.

IBM says it is the fastest quantum chip it has built. It runs accurate circuits with more than 7,500 quantum operations. A circuit is just a quantum program, written as a list of steps.

• Speed: up to 100,000 samples per second.
• Gain: 25 times more than IBM's Heron chip.
• Source of the speed: a fast reset that clears the qubits between runs.

In real life Think of a sprinter who can get back to the starting line almost instantly. The race is the same length, but you fit in far more races per hour.

IBM says partners including E.ON, Algorithmiq and Q-CTRL have run more than 34 billion circuits on Nighthawk r2. They report gains of up to 19 times over earlier IBM chips. The 34 billion count comes from IBM's own team calculations.

IBM also described a run with Oak Ridge National Laboratory. It simulated real magnetic materials and got a 12 times speedup. The output matched experiments that fire neutrons at materials to see how they behave.

Another case came from Algorithmiq, which simulated how a mixed material changes over time. On the new chip, the quantum part ran 8 times faster. Classical methods would need an estimated 240 times more memory to keep up at the same accuracy.

Everyday quantum chips make mistakes. A fault-tolerant computer fixes its own mistakes as it goes, so it can finish very long programs. IBM Quantum Starling is the one it plans to deliver in 2029. It aims for circuits with 100 million operations.

IBM works on the parts in parallel, because a lesson on one piece changes the design of the others. Here is where it says each stands:

• Loon chip: it puts the ingredients of an error-correcting design on one chip. IBM cut its hardest-to-build parts, called Josephson junctions, from about 5,000 to 2,000.
• Chip-to-chip links: two Nighthawk chips were joined with four links, called l-couplers. They passed quantum states at 99% average accuracy, with a best result of 99.3%.
• Decoders: these are programs that spot errors in real time. IBM says they now meet what Starling needs, and it is building a dedicated decoder chip.
• Control electronics: IBM says its newest cold-running control chips match the gate performance of room-temperature electronics.
• Cooling: IBM joined two cooling modules into one system for the first time. It reached operating temperature in under five days and cooled below 15 millikelvin, a tiny fraction of a degree above absolute zero.

Starling needs more than 100 of those chip-to-chip links, so the 99% result is an early proof point rather than a finish.

IBM says the future is not one circuit at a time. Real jobs need hundreds of millions, even billions, of samples spread across quantum and classical computers. Its answer is a "directed execution model." Instead of a black box that runs one circuit and returns an answer, researchers get more control and a clearer view of the whole run.

IBM also named new tools:

• Profiling tools: estimate a job's cost and timing before you run it, and check noise on your own computer.
• Backend v3: coming in 2027, it treats the classical steps before and after the quantum run as one program.
• Qiskit Fermions and Qiskit Bosons: new application tools, plus optimization mappers.
• Agent skills: an open-source set of packaged instructions that help an AI coding assistant follow current Qiskit best practices.

Qiskit is IBM's open-source toolkit for quantum programming. Open-source means anyone can read, use and change its code.

Qubits are delicate. A little heat or noise ruins a calculation, so every serious plan leans on error correction: ways to detect and repair mistakes while a program runs. IBM's developer guide calls it one of the biggest themes of the event.

IBM says users can now run a repetition code, a basic error-correcting test, for 1,000,000 rounds in a single circuit. Each round measures qubits in the middle of the program and resets them. A new loop instruction repeats those steps without writing out a gigantic circuit.

IBM labels some of these features experimental, and they may change. IBM is also putting Nighthawk r2 next to Oak Ridge's Umbra supercomputer, which runs on AMD chips. The goal is to design error-correction methods across quantum and classical machines together. This work fits the US Department of Energy's Genesis Mission, which aims for 100 logical qubits running 100,000 hard operations by late 2028.

It is not only a lab story. IBM announced on November 6 that a Quantum System Two with Nighthawk is running at Hyde Park Labs in Chicago, its first quantum computer in Illinois. Under a plan announced in December 2024, that machine was to use the Heron chip instead.

Nighthawk has fewer qubits than the 156 in Heron r2 and r3, but more couplers, the parts that link qubits. The count rose from up to 176 to 218. The speed numbers are IBM's own specifications, not measured on the Chicago unit.

• If you code: you can run circuits on IBM Quantum Platform today, and IBM has published hands-on tutorials and expanded free access.
• If you follow tech: treat "advantage" as a first step. The big promise needs Starling, which is still targeted for 2029.
• If you work in chemistry, materials or finance: IBM lists these fields as targets, with 39 client projects already testing the new application tools.
• If you are just curious: nothing in your phone or laptop changes this year.