A new proposal from top researchers at Microsoft and Qolab aims to replace headline-grabbing quantum claims with a common standard for measuring progress toward useful quantum computers by establishing a workable definition of scalable logical qubits. Many quantum announcements describe breakthroughs in terms of logical qubits. These are error-protected units of quantum information needed to build useful quantum machines. Unfortunately, without a standard definition for them, it can be difficult to determine the validity of a vendor’s claims.

There are a number of potential disconnects. For example, some demonstrations detect errors but cannot correct them, while others run for only a handful of cycles — not long enough to be useful in production. There are still others that support only a subset of the operations needed by quantum computers. Partly because companies have an incentive to present their results in the best light, “logical qubits” has come to mean many different things. This is why I argued, earlier this year, that the quantum computing industry needs logical qubit standards . (Note: Microsoft, Google and AWS are advisory clients of my firm, Moor Insights & Strategy. Quantinuum and Atom Computing, mentioned below, are past clients.)

Help may be on the way. In a new preprint on arXiv, “ Scalable Logical Qubits ,” three eminent quantum scientists propose a new framework for assessing quantum progress, one that I believe should be useful to researchers, investors and policymakers by helping them distinguish between actual quantum developments and one-off demos.

The State Of Qubits Today And How It Needs To Evolve

As a reminder, smaller numbers of logical qubits are formed from much larger numbers of physical qubits. Even today’s best physical qubits typically introduce errors at a rate of roughly one in every thousand operations. That is the result of physical qubits being affected by heat, stray electromagnetic signals and tiny manufacturing imperfections. But while that might sound like a small number of errors, in fact it’s much too high for ongoing practical uses. The authors expect the low end of useful, utility-scale applications, running on 100 or more logical qubits — and therefore many times that number of physical qubits — to require error rates of about 1 in 10 billion. Larger applications, such as chemistry simulations or code-breaking algorithms, will require something closer to one error in 1 quadrillion operations.

Physical hardware improvements alone cannot close this gap. As software programs grow longer, the chance of an error-free run shrinks exponentially. To counter this, physical error rates must be pushed below the fault-tolerance threshold, and active quantum error correction must run throughout the computation. That is why the authors have proposed the concept of “scalable logical qubits.” They argue that the design of a scalable logical qubit must meet the following four conditions:

  1. It must be continuously sustained through repeated error correction during the computation.
  2. It must support a universal set of operations that are performed in an error-protected way. Programs must also be able to change course mid-computation based on measurement results, which requires low-latency, real-time decoding.
  3. It must be an instance of a code family in which the logical error rate decreases predictably as more physical qubits are devoted to it.
  4. There must be a credible path to replicate it hundreds or thousands of times. That must be done without degrading the logical qubit’s performance, and without the number of physical qubits ballooning. (If you’re interested in knowing the math for this, the authors cap it at roughly the number of logical qubits multiplied by the logarithm of the inverse target error rate.)

The paper explains why error correction needs to run far more cycles than current demonstrations allow. It also describes a complete set of capabilities that must be shown, particularly magic-state-based universal gates and real-time feedback. Logical error rates must fall predictably toward the 10^-12 to 10^-15 range as more physical qubits are devoted to each logical qubit. Systems must be able to grow from a handful of logical qubits to hundreds or thousands without increasing errors, especially correlated errors that strike many qubits at once and are notoriously hard to correct. Wiring, electronics and classical control systems must also be engineered for utility-scale machines, possibly through modular or networked designs.

Rather than ranking systems with a single metric, the authors argue that scalable logical qubits should be measured across four linked dimensions:

  • Reliability — A scalable logical qubit must outperform its underlying physical qubits under repeated error correction. It must also have a path to achieve error rates of one in a trillion to one in a quadrillion.
  • Scale — For practical implementation, we need to know how many logical qubits can operate simultaneously, with no degradation in quality as more are added.
  • Capability — Qubits must support state preparation and all of the following: measurement, memory, fault-tolerant Clifford gates, universal non-Clifford gates (typically via magic-state preparation and distillation) and measurement-conditioned control flow.
  • Performance — A scalable logical qubit must have fast logical cycles and efficient gate synthesis that maintain practical wall-clock runtime and operational costs at scale.

These four dimensions are in a tug-of-war with one another. Consider how a developer must operate within a fixed budget of physical qubits. That developer can use that finite inventory to build either many mediocre logical qubits or only a few good ones. It is worth considering that some error-correcting codes may reduce qubit count, potentially resulting in longer operations. Likewise, some decoders may achieve higher accuracy, but with the tradeoffs of higher classical computational cost and increased decoding latency, which slows execution. Finally, more aggressive compilation can trim a program’s runtime or qubit needs, but it shifts the burden to classical computing, either before the program runs or inside the real-time control loop.

The paper also discusses techniques that complement error correction. In my view, these are also the techniques most likely to flatter a logical qubit announcement. Improving physical-qubit fidelity remains foundational, because it lowers the overhead for error correction. One complementary technique is post-selection, in which operational runs are discarded after an error is detected. Another is error mitigation, which statistically corrects results after the fact rather than during the computation, but at the cost of extra sampling runs.

Both post-selection and error mitigation have real uses, and the authors describe them as valuable complements. However, neither can substitute for error correction itself in long computations. With post-selection, the probability of a flawless run shrinks as programs grow, and mitigation’s sampling overhead and underlying assumptions become harder to sustain at large scale.

The Quantum Computing Cost-Benefit Reality Check

The authors argue that once systems approach application scale, all of these concepts ultimately reduce to a single, straightforward economic consideration: the value derived from the resources put into the system. This is a standard that DARPA applies when evaluating advanced systems. And we will be able to declare that utility-scale quantum computing has been achieved at the point where the value of a computation exceeds the cost of performing it.

Total cost must include runtime, the number of physical qubits per logical qubit, the classical computing power required to correct errors and the number of times a calculation must be repeated. This is why the raw number of logical qubits in a quantum system is not meaningful on its own — another reminder to take company announcements with a grain of salt. What matters is how reliably a system can compute per unit of time and at what price.

To be fair, we have every reason to believe there will be major payoffs for achieving error-corrected quantum computing goals. There is great value, for example, in the ability to simulate molecules and materials to design drugs, catalysts, batteries and superconductors. There is likewise high value in mapping chemical reaction pathways and modeling physical systems beyond the reach of classical supercomputers. These are the benefits that should, someday, justify the costs.

An Unusual Cross-Company Collaboration

This paper also has particular value for influencing discussion in the quantum industry because of its authorship. It was written by three highly respected quantum scientists: Matthias Troyer , Chetan Nayak and John Martinis . Troyer is a Microsoft Technical Fellow as well as the company’s vice president of quantum. Formerly a professor of computational physics at ETH Zurich, he pioneered algorithms for simulating quantum many-body systems and led the open-source ALPS project. Nayak, also a Microsoft Technical Fellow, is a professor of physics at the University of California, Santa Barbara. He has worked on topological phases, high-temperature superconductivity, the quantum Hall effect, time crystals, non-Abelian anyons and Majorana zero modes. As I discussed in a recent article, he leads Microsoft’s development program for Majorana quantum chips .

While Drs. Troyer and Nayak have each earned a slew of scientific honors, Dr. Martinis might be the best-known of the three, especially after he and two colleagues were awarded the 2025 Nobel Prize in Physics. Like Nayak, Martinis is a professor at UCSB. He is known for his pioneering work in quantum computing and the discovery of macroscopic quantum phenomena in superconductors. He also led the Google Quantum AI Lab team when it used the Sycamore processor — which his team designed — to claim the first demonstration of quantum supremacy in 2019. Martinis recently cofounded Qolab, a superconducting startup.

Note that two of the authors are from Microsoft, which is pursuing topological qubits and has partnered with trapped-ion and neutral-atom companies. Meanwhile, Martinis had a long association with Google and is one of the great proponents of superconducting for quantum computing. In my view, a hardware-agnostic framework endorsed by scientists from competing hardware approaches (and competing tech giants) ought to carry more weight than one coming from a single company. Whether their ideas and their new definition of scalable logical qubits gain broad support will depend on the quantum community’s willingness to test and adopt their suggestions.

How To Read The Next Logical Qubit Announcement

The three scientists’ paper closes with a practical checklist that readers can apply to any logical qubit claim. The questions include:

  • How many cycles of repeated error correction were demonstrated, and what would it take to run more?
  • Which capabilities were shown: state preparation, memory, Clifford gates, universal gates, real-time feedback?
  • Are logical error rates better than physical ones, and do they fall as more physical qubits are added to each logical qubit?
  • How many logical qubits were realized, and can that number grow without raising error rates?
  • What control architecture would support the number and quality of qubits that useful applications require?

An announcement that answers these questions clearly deserves more attention than one built around a single headline number.

It is important to mention that teams using superconducting, trapped-ion, neutral-atom and bosonic-qubit approaches, including from Google, AWS, Quantinuum, Atom Computing and Harvard/QuEra, have all recently reported impressive experimental results based on logical qubits. Several of the demonstrations the paper cites involved Microsoft as a collaborator.

Beyond these identified milestones lie further trade-offs in the future. We will need faster and more accurate decoders as more efficient codes emerge. The field has already made great progress with qLDPC codes and smarter compilers that reduce the resources programs need. There must also be continued improvements in the physical qubits themselves.

Even so, the authors have provided a yardstick with great potential. If the ideas in this paper are shared and adopted broadly, then the next wave of breakthrough announcements in quantum will likely be easier to judge and implement.