Debunking algorithmic qubits
Executive Summary: ĢƵ’s H-Series computers have the highest performance in the industry, verified by multiple widely adopted benchmarks including quantum volume We demonstrate that an alternative benchmark called algorithmic qubits is deeply flawed, hiding computer performance behind a plurality voting trick and gate compilations that are not widely useful.
Recently a new benchmark called algorithmic qubits (AQ) has started to be confused with quantum volume measurements. Quantum volume (QV) was specifically designed to be hard to “game,” however the algorithmic qubits test turns out to be very susceptible to tricks that can make a quantum computer look much better than it actually is. While it is not clear what can be done to fix the algorithmic qubits test, it is already clear that it is much easier to pass than QV and is a poor substitute for measuring performance. It is also important to note that algorithmic qubits are not the same as logical qubits, which are necessary for full fault-tolerant quantum computing.
To make this point clear, we simulated what algorithmic qubits data would look like for two machines, one clearly much higher performing than the other. We applied two tricks that are typically used when sharing algorithmic qubits results: gate compilation and . From the data above, you can see how these tricks are misleading without further information. For example, if you compare data from the higher fidelity machine without any compilation or plurality voting (bottom left) to data from the inferior machine with both tricks (top right) you may incorrectly believe the inferior machine is performing better. Unfortunately, this inaccurate and misleading comparison has been made in the past. It is important to note that algorithmic qubits uses a subset of algorithms from a that introduced a suite of application oriented tests and created a repository to test available quantum computers. Importantly, that work explicitly forbids the compilation and error mitigation techniques that are causing the issue here.
As a demonstration of the perils of AQ as a benchmark, we look at data obtained on both ĢƵ’s H2-1 system as well as publicly available data from IonQ’s Forte system.
We reproduce data without any error mitigation from IonQ’s in association with a preprint posted to the , and compare it to data taken on our H2-1 device. Without error mitigation, IonQ Forte achieves an AQ score of 9, whereas ĢƵ H2-1 achieves AQ of 26. Here you can clearly see improved circuit fidelities on the H2-1 device, as one would expect from the higher reported 2Q gate fidelities (average 99.816(5)% for ĢƵ’s H2-1 vs 99.35% for IonQ’s Forte). However, after you apply error mitigation, in this case plurality voting, to both sets of data the picture changes substantially, hiding each underlying computer’s true capabilities.
Here the H2-1 algorithmic performance still exceeds Forte (from the publicly released data), but the perceived gap has been reduced by error mitigation.
“Error mitigation, including plurality voting, may be a useful tool for some near-term quantum computing but it doesn’t work for every problem and it’s unlikely to be scalable to larger systems. In order to achieve the lofty goals of quantum computing we’ll need serious device performance upgrades. If we allow error mitigation in benchmarking it will conflate the error mitigation with the underlying device performance. This will make it hard for users to appreciate actual device improvements that translate to all applications and larger problems,” explained Dr. Charlie Baldwin, a leader in ĢƵ’s benchmarking efforts.
There are other issues with the algorithmic qubits test. The circuits used in the test can be reduced to very easy-to-run circuits with basic quantum circuit compilation that are freely available in packages like . For example, the largest phase estimation and amplitude estimation tests required to pass AQ=32 are specified with 992 and 868 entangling gates respectively but applying pytket optimization reduces the circuits to 141 and 72 entangling gates. This is only possible due to choices in constructing the benchmarks and will not be universally available when using the algorithms in applications. Since AQ reports the precompiled gate counts this also may lead users to expect a machine to be able to run many more entangling gates than what is actually possible on the benchmarked hardware.
What makes a good quantum benchmark? Quantum benchmarking is extremely useful in charting the hardware progress and providing roadmaps for future development. However, quantum benchmarking is an evolving field that is still an open area of research. At ĢƵ we believe in testing the limits of our machine with a variety of different benchmarks to learn as much as possible about the errors present in our system and how they affect different circuits. We are open to working with the larger community on refining benchmarks and creating new ones as the field evolves.
To learn more about the Algorithmic Qubits benchmark and the issues with it, please watch this video where Dr. Charlie Baldwin walks us through the details, starting at 32:40.
About ĢƵ
ĢƵ, the world’s largest integrated quantum company, pioneers powerful quantum computers and advanced software solutions. ĢƵ’s technology drives breakthroughs in materials discovery, cybersecurity, and next-gen quantum AI. With over 500 employees, including 370+ scientists and engineers, ĢƵ leads the quantum computing revolution across continents.
This month, ĢƵ welcomed its global user community to the first-ever Q-Net Connect, an annual forum designed to spark collaboration, share insights, and accelerate innovation across our full-stack quantum computing platforms. Over two days, users came together not only to learn from one another, but to build the relationships and momentum that we believe will help define the next chapter of quantum computing.
Q-Net Connect 2026 drew over 170 attendees from around the world to Denver, Colorado, including representatives from commercial enterprises and startups, academia and research institutions, and the public sector and non-profits - all users of ĢƵ systems.
The program was packed with inspiring keynotes, technical tracks, and customer presentations. Attendees heard from leaders at ĢƵ, as well as our partners at NVIDIA, JPMorganChase and BlueQubit; professors from the University of New Mexico, the University of Nottingham and Harvard University; national labs, including NIST, Oak Ridge National Laboratory, Sandia National Laboratories and Los Alamos National Laboratory; and other distinguished guests from across the global quantum ecosystem.
Congratulations to Q-Net Connect 2026 Award Recipients!
The mission of the ĢƵ Q-Net user community is to create a space for shared learning, collaboration and connection for those who adopt ĢƵ’s hardware, software and middleware platform. At this year’s Q-Net Connect, we awarded four organizations who made notable efforts to champion this effort.
- JPMorganChase received the ‘Guppy Adopter Award’ for their exemplary adoption of our quantum programming language, Guppy, in their research workflows.
- Phasecraft, a UK and US-based quantum algorithms startup, received the ‘Rising Star’ award for demonstrating exceptional early impact and advancing science using ĢƵ hardware, which they published in a December 2025 .
- Qedma, a quantum software startup, received the ‘Startup Partner Engagement’ award for their sustained engagement with ĢƵ platforms dating back to our first commercially deployed quantum computer, H1.
- Anna Dalmasso from the University of Nottingham received our ‘New Student Award’ for her impressive debut project on ĢƵ hardware and for delivering outstanding results as a new Q-Net student user.
Congratulations, again, and thank you to everyone who contributed to the success of the first Q-Net Connect!
Become a Q-Net Member
Q-Net offers year‑round support through user access, developer tools, documentation, trainings, webinars, and events. Members enjoy many exclusive benefits, including being the first to hear about exclusive content, publications and promotional offers.
By joining the community, you will be invited to exclusive gatherings to hear about the latest breakthroughs and connect with industry experts driving quantum innovation. Members also get access to Q‑Net Connect recordings and stay connected for future community updates.
In a follow-up to our recent work with Hiverge using AI to discover algorithms for quantum chemistry, we’ve teamed up with Hiverge, Amazon Web Services (AWS) and NVIDIA to explore using AI to improve algorithms for combinatorial optimization.
With the rapid rise of Large Language Models (LLMs), people started asking “what if AI agents can serve as on-demand algorithm factories?” We have been working with Hiverge, an algorithm discovery company, AWS, and NVIDIA, to explore how LLMs can accelerate quantum computing research.
Hiverge – named for Hive, an AI that can develop algorithms – aims to make quantum algorithm design more accessible to researchers by translating high-level problem descriptions in mostly natural language into executable quantum circuits. The Hive takes the researcher’s initial sketch of an algorithm, as well as special constraints the researcher enumerates, and evolves it to a new algorithm that better meets the researcher’s needs. The output is expressed in terms of a familiar programming language, like Guppy or , making it particularly easy to implement.
The AI is called a “Hive” because it is a collective of LLM agents, all of whom are editing the same codebase. In this work, the Hive was made up of LLM powerhouses such as Gemini, ChatGPT, Claude, Llama, as well as which was accessed through AWS’ Amazon Bedrock service. Many models are included because researchers know that diversity is a strength – just like a team of human researchers working in a group, a variety of perspectives often leads to the strongest result.
Once the LLMs are assembled, the Hive calls on them to do the work writing the desired algorithm; no new training is required. The algorithms are then executed and their ‘fitness’ (how well they solve the problem) is measured. Unfit programs do not survive, while the fittest ones evolve to the next generation. This process repeats, much like the evolutionary process of nature itself.
After evolution, the fittest algorithm is selected by the researchers and tested on other instances of the problem. This is a crucial step as the researchers want to understand how well it can generalize.
In this most recent work, the joint team explored how AI can assist in the discovery of heuristic quantum optimization algorithms, a class of algorithms aimed at improving efficiency across critical workstreams. These span challenges like optimal power grid dispatch and storage placement, arranging fuel inside nuclear reactors, and molecular design and reaction pathway optimization in drug, material, and chemical discovery—where solutions could translate into maximizing operational efficiency, dramatic reduction in costs, and rapid acceleration in innovation.
In other AI approaches, such as reinforcement learning, models are trained to solve a problem, but the resulting "algorithm" is effectively ‘hidden’ within a neural network. Here, the algorithm is written in Guppy or CUDA-Q (or Python), making it human-interpretable and easier to deploy on new problem instances.
This work leveraged the NVIDIA CUDA-Q platform, running on powerful NVIDIA GPUs made accessible by AWS. It’s state-of-the art accelerated computing was crucial; the research explored highly complex problems, challenges that lie at the edge of classical computing capacity. Before running anything on ĢƵ’s quantum computer, the researchers first used NVIDIA accelerated computing to simulate the quantum algorithms and assess their fitness. Once a promising algorithm is discovered, it could then be deployed on quantum hardware, creating an exciting new approach for scaling quantum algorithm design.
More broadly, this work points to one of many ways in which classical compute, AI, and quantum computing are most powerful in symbiosis. AI can be used to improve quantum, as demonstrated here, just as quantum can be used to extend AI. Looking ahead, we envision AI evolving programs that express a combination of algorithmic primitives, much like human mathematicians, such as Peter Shor and Lov Grover, have done. After all, both humans and AI can learn from each other.
As quantum computing power grows, so does the difficulty of error correction. Meeting that demand requires tight integration with high-performance classical computing, which is why we’ve partnered with NVIDIA to push the boundaries of real-time decoding performance.
Realizing the full power of quantum computing requires more than just qubits, it requires error rates low enough to run meaningful algorithms at scale. Physical qubits are sensitive to noise, which limits their capacity to handle calculations beyond a certain scale. To move beyond these limits, physical qubits must be combined into logical qubits, with errors continuously detected and corrected in real time before they can propagate and corrupt the calculation. This approach, known as fault tolerance, is a foundational requirement for any quantum computer intended to solve problems of real-world significance.
Part of the challenge of fault tolerance is the computational complexity of correcting errors in real time. Doing so involves sending the error syndrome data to a classical co-processor, solving a complex mathematical problem on that processor, then sending the resulting correction back to the quantum processor - all fast enough that it doesn’t slow down the quantum computation. For this reason, Quantum Error Correction (QEC) is currently one of the most demanding use-cases for tight coupling between classical and quantum computing.
Given the difficulty of the task, we have partnered with NVIDIA, leaders in accelerated computing. With the help of NVIDIA’s ultra-fast GPUs (and the GPU-accelerated BP-OSD decoder developed by NVIDIA as part of library), we were able to demonstrate real-time decoding of Helios’ qubits, all in a system that can be connected directly to our quantum processors using .
While real-time decoding has been demonstrated before (notably, by our own scientists in this study), previous demonstrations were limited in their scalability and complexity.
In this demonstration, we used Brings’ code, a high-rate code that is possible with our all-to-all connectivity, to encode our physical qubits into noise-resilient logical qubits. Once we had them encoded, we ran gates as well as let them idle to see if we could catch and correct errors quickly and efficiently. We submitted the circuits via both as well as our own Guppy language, underlining our commitment to accessible, ecosystem-friendly quantum computing.
The results were excellent: we were able to perform low-latency decoding that returned results in the time we needed, even for the faster clock cycles that we expect in future generation machines.
A key part of the achievement here is that we performed something called “correlated” decoding. In correlated decoding, you offload work that would normally be performed on the QPU onto the classical decoder. This is because, in ‘standard’ decoding, as you improve your error correction capabilities, it takes more and more time on the QPU. Correlated decoding elides this cost, saving QPU time for the tasks that only the quantum computer can do.
Stay tuned for our forthcoming paper with all the details.