Better benchmarks, closer scrutiny, and advances in error correction are giving us more useful ways to judge quantum progress.
Quantum computing is approaching a test that every technology eventually has to pass: can it do useful work reliably enough to justify what it costs?
For IT professionals, that question should sound familiar. An impressive demonstration can earn our attention. Putting a system into service requires evidence about performance, reliability, operating requirements, and the problem it actually solves.
That is why a recent announcement from the Defense Advanced Research Projects Agency deserves attention. On October 7, DARPA announced that Atom Computing, Diraq, IBM, and IonQ had advanced to Stage C, the final stage of its Quantum Benchmarking Initiative. They join Microsoft and PsiQuantum, which entered through a predecessor program. [1]
The initiative is examining whether quantum computing approaches can deliver utility-scale operation by 2033, defined in terms of computational value exceeding cost. Stage C brings government evaluators alongside developers to scrutinize whether proposed systems can be built and operated as intended. [1]
Advancing to that stage does not establish that a company has achieved useful quantum computing. It means its approach has reached a deeper level of evaluation.
That is exactly what makes the announcement worth discussing.
The question behind the announcements
Quantum computers use quantum states to process information in ways that could make certain calculations substantially more tractable. Potential applications include simulating molecules and materials, where quantum behavior is part of the problem itself.
However, a potential advantage has to survive contact with the full workload. Preparing the input, running the calculation, managing errors, and obtaining a useful answer all matter. So does comparison with the strongest available classical method.
The practical question is specific: which problem can this system solve, with what accuracy, in how much time, and at what cost?
Those questions give both researchers and business leaders something more useful than a headline about a faster machine.
Counting qubits only tells part of the story
Qubit counts are easy to put in a headline. Understanding what those qubits can accomplish takes more work.
Physical qubits are the underlying hardware elements. They are also vulnerable to errors. Logical qubits encode information across physical qubits so that suitable errors can be detected and corrected.
This matters because a useful calculation may require a long sequence of operations. A machine needs to preserve the information well enough to reach a trustworthy result.
Quantinuum reported in March 2026 that its Helios system supported demonstrations involving 48 error-corrected logical qubits using 98 physical qubits. That is an important engineering result. It also needs to be read alongside the details of the encoding, supported operations, and measured performance. [2]
Two systems with the same logical-qubit count may support different workloads and levels of reliability. The number alone does not settle which system is more capable.
For anyone who has evaluated servers, storage, or networks, this is familiar territory. A specification becomes useful when you understand the workload and the conditions under which it was measured.
Reliability is becoming the central engineering story
Error correction deserves more attention than it usually gets outside the research community.
The challenge is to detect and correct errors while a computation continues, quickly enough and with sufficiently little overhead to support useful work. Adding hardware helps only if the resulting system can perform a larger, more reliable calculation.
A September research preprint titled Scalable logical qubits proposes evaluating progress across reliability, scale, capability, and performance. Its authors emphasize repeated error correction, fault-tolerant operations, and timely decoding and feedback. [3]
That is a valuable way to frame the field. Can the system protect information? Can it perform the operations an application needs? Can that performance be sustained and expanded?
These are difficult engineering questions. Progress toward answering them is a reason for excitement.
Better benchmarks can make progress easier to judge
Another September preprint introduced QUOPS, a framework for measuring the size of computationally relevant circuits a quantum processor can successfully execute and how quickly it can execute them. The researchers applied it to processors from Quantinuum, Google, and IBM. [4]
For the useful-computation challenge problems examined in that study, the authors estimated that computational capability would need to grow by roughly five orders of magnitude. [4]
That estimate is specific to their methodology and selected problems. It should not be treated as a universal countdown to quantum usefulness.
Its value is that it connects present performance to explicit computational requirements. It helps turn a broad promise into something researchers can measure, challenge, and improve.
The paper is a preprint, so its findings should be read as research subject to further scrutiny. Nevertheless, the effort to compare different systems against a meaningful task is an encouraging development.
A roadmap creates a target we can evaluate
IBM has set a 2029 target for Starling, a fault-tolerant quantum system intended to support 200 logical qubits and circuits containing 100 million quantum gates. Its roadmap describes the error-correction and systems work needed to reach that goal. [5]
A target of that specificity gives the public something concrete to follow. It also remains a target. Announcing an architecture and demonstrating its intended performance are separate milestones.
The same discipline should apply across vendors. Ask what has been demonstrated, what assumptions remain, and what evidence would establish the next milestone.
DARPA explicitly says its initiative is not intended to select a single winner. Different hardware approaches are being evaluated on their own merits. [1] That is sensible in a field where several major engineering paths remain under development.
Classical computing keeps improving too
There is another reason quantum claims need continuing evaluation: the systems used for comparison keep getting better.
On October 15, researchers at the University of Edinburgh and EPCC announced a more efficient algorithm for simulating certain photonic quantum systems. They reported reducing the computing resources needed to around five percent of the previous requirement, bringing some simulations within reach of a powerful desktop. Their work was published in Science Advances. [6]
The scope matters. This concerns Gaussian boson sampling simulations, rather than a general ability to reproduce any quantum computer on a desktop.
Still, the lesson is significant. Better classical algorithms can change the comparison against which quantum advantage is claimed. They can also help researchers design and understand quantum hardware at lower cost.
Progress on one side can accelerate research on the other.
What should we take from this?
For technology leaders, the useful response is to follow the evidence and identify where it could affect their organization.
When evaluating a quantum announcement, ask:
- What was actually demonstrated, and what remains a forecast?
- What problem did the system solve, and how was the answer checked?
- Was it compared with a strong, current classical approach?
- What resources were required for the complete calculation?
- Can the result be reproduced and extended to a useful workload?
These questions support better investment decisions and more accurate reporting. They also make technical achievements easier to appreciate: a result is more impressive when we understand precisely what it establishes.
For cybersecurity teams, another workstream deserves attention alongside hardware progress. IBM notes that identifying and migrating vulnerable cryptography can take years and recommends planning for cryptographic agility and the risk of information being collected now for later decryption. [7]
That work can begin with understanding where cryptography is used, who owns it, and how it can be changed. A hardware roadmap should inform that planning without being mistaken for a firm date when existing encryption will fail.
Why this deserves attention now
The encouraging development is that quantum progress is becoming more open to practical evaluation.
Independent scrutiny can expose weak assumptions early. Better benchmarks can show where a system is improving. Error-correction research can establish whether longer and more useful calculations are becoming feasible.
I find that more compelling than another isolated qubit record. It brings the discussion closer to the questions that determine whether technology can earn a place in real operations.
We should expect ambitious claims from a field pursuing difficult breakthroughs. We should also expect evidence that lets us assess them.
Quantum computing has plenty of promise. The next stage is showing, in reproducible and economically meaningful terms, what that promise can deliver.
Sources
- DARPA, October 7, 2026 – Four more teams enter Quantum Benchmarking Initiative’s final stage.
- Quantinuum, March 4, 2026 – Skinny Logic Quantum Codes Go on a Diet. Vendor research announcement.
- Troyer, Nayak, and Martinis, September 17, 2026 – Scalable logical qubits. Research preprint.
- Proctor and colleagues, September 10, 2026 – Benchmarking the computational power of quantum computers. Research preprint.
- IBM, June 10, 2025 – How IBM will build the world’s first large-scale fault-tolerant quantum computer. Vendor roadmap.
- EPCC, October 15, 2026 – Software breakthrough could spur quantum computer advances. Institutional summary linking to the Science Advances paper.
- IBM – Quantum 2026 Technology Atlas. Roadmap updated March 2026.
