Quantinuum, IonQ, IBM, Google: who is ahead in quantum?
Every funding round, every investor, and who is really ahead in quantum computing, in one document. Have a look ↓
Summary
Quantinuum is now the clearest leader on the layer that decides fault-tolerant quantum computing: logical qubits that make fewer errors than the hardware underneath. Google still leads on verifiable quantum advantage, IBM on industrial roadmap, and IonQ on revenue and peak physical fidelity.
The race still has no single winner, but the balance has moved. In October 2026, Quantinuum said its Helios machine reached 99.9995% logical memory fidelity with a distance-7 color code, about five logical errors per million error-correction rounds. The company says that is roughly 100 times better than Google’s best published superconducting result. The figure comes from the company and a research preprint, not yet a peer-reviewed paper.
On a simple 23-criteria yes-count, Quantinuum comes first with 17 positive checks. IBM and IonQ tie at 11, and Google has 10. Read this as breadth of public evidence, not a weighted technical verdict. Google still ranks second because our weights favor the hardest milestones, and it has passed several of them.
Qubit count is no longer the metric that decides the race. The better question is how many reliable logical operations a machine can run before noise destroys the computation.
Google’s case is still the deepest scientifically. Willow’s below-threshold error correction, the Quantum Echoes advantage claim, real-time decoding and, in October 2026, a Science paper on magic state cultivation with 99.99% T-state fidelity form the strongest proof package outside Quantinuum.
Quantinuum’s case now combines proof and money. It listed on Nasdaq in June 2026, raising $1.68 billion at $60 a share, a reported valuation of roughly $14 billion to $15 billion at the IPO price. The shares opened at $68 but traded near $35 in October 2026. It reported $8 million of revenue in the second quarter of 2026 and guides $28 million to $32 million for the year.
IBM is closing part of the gap on logical computing. In July 2026, IBM and the University of Chicago reported a run on 70 error-detected logical qubits that they call verifiable quantum advantage; it is still a preprint. IBM looks much stronger on manufacturing, dated targets and outside validation: in October 2026, DARPA moved IBM to Stage C of its Quantum Benchmarking Initiative, where hardware gets tested by an independent team.
IonQ is the commercial leader by far. It reported $80.1 million of revenue in the second quarter of 2026, closed its SkyWater foundry deal in July and also reached DARPA Stage C. But its public logical-qubit evidence on real hardware still trails Quantinuum and Google.
The industry is transparent enough to compare directionally, not enough to score everyone on one clean sheet. Each company pushes the metric where it looks best, so the comparison has to separate physics, logical execution, customer access and roadmap credibility.
The most overread numbers are probably IonQ’s 99.99% physical fidelity and its revenue, IBM’s 2029 roadmap, and Google’s RCS timing headline. They matter, but none proves useful fault-tolerant computing today.
Our verdict: Quantinuum is ahead overall on the evidence that counts most, Google is a close second with the strongest advantage proof, IBM has the most credible industrial plan, and IonQ sells the most but still has to show logical computing on its own hardware.
Written by the New Market Pitch teamWe track new markets so founders and investors can move faster. About us
We keep an up-to-date report on quantum computing. Have a look ↓
Sample pages from an earlier edition. The report has been updated with new data since.
As of today, who is ahead in quantum computing?
As of today, Quantinuum is ahead in logical-qubit and error-correction execution, Google is ahead in demonstrated quantum advantage, IBM is ahead in industrial roadmap, and IonQ is ahead in revenue and peak physical fidelity but less proven on logical computing.
What the lead rests on matters. Quantinuum’s lead rests on published logical-qubit results on a commercial machine, not on its IPO. IonQ’s revenue is ten times Quantinuum’s, but revenue measures sales of today’s noisy machines, not progress toward fault tolerance. IBM and IonQ entering DARPA’s Stage C in October 2026 is a real vote of confidence, but it opens a hardware test; it does not report one.
If we judge them on 23 criteria, here is the yes-count ranking we get: Quantinuum 17, IBM 11, IonQ 11, Google 10. Quantinuum’s Apollo system, described as universal and fully fault-tolerant, is an explicit 2029 target that the company says is on schedule. IBM’s July 2026 experiment gives it multiple logical qubits and logical two-qubit gates on hardware, though with error detection rather than full correction.
Here are the details:
| # | Question | Quantinuum | IonQ | IBM | |
|---|---|---|---|---|---|
| 1 | Has working quantum processors today? | yes | yes | yes | yes |
| 2 | Publishes public hardware benchmarks? | yes | yes | yes | yes |
| 3 | Maintains a public quantum software stack? | yes | yes | yes | yes |
| 4 | Offers customer cloud access today? | yes | yes | yes | unclear |
| 5 | Shows 50+ qubit-scale capability? | yes | yes | yes | yes |
| 6 | Has a broad quantum user ecosystem? | yes | yes | yes | unclear |
| 7 | Has full-stack quantum control capability? | yes | yes | yes | yes |
| 8 | Demonstrates all-to-all qubit connectivity? | yes | yes | no | no |
| 9 | Reports 99.9% two-qubit fidelity? | yes | yes | unclear | unclear |
| 10 | Reports 99.99% two-qubit fidelity? | no | yes | no | no |
| 11 | Demonstrated beyond-classical sampling? | yes | unclear | unclear | yes |
| 12 | Demonstrated verifiable quantum advantage? | no | no | unclear | yes |
| 13 | Demonstrated real-time error correction? | yes | unclear | unclear | unclear |
| 14 | Demonstrated logical qubits beating physical? | yes | unclear | unclear | yes |
| 15 | Demonstrated repeated error correction cycles? | yes | unclear | unclear | yes |
| 16 | Demonstrated multiple logical qubits? | yes | unclear | yes | unclear |
| 17 | Demonstrated logical two-qubit gates? | yes | unclear | yes | unclear |
| 18 | Demonstrated scalable real-time decoding? | yes | unclear | unclear | yes |
| 19 | Published end-to-end FTQC architecture? | unclear | yes | yes | unclear |
| 20 | Has explicit 2029 FTQC target? | yes | no | yes | no |
| 21 | Delivered useful fault-tolerant applications? | no | no | no | no |
| 22 | Runs million-gate logical circuits today? | no | no | no | no |
| 23 | Has commercial fault-tolerant quantum today? | no | no | no | no |
If you want more recent data on this point, see the latest figures in our Quantum Computing report ↓.
These pages are excerpts from our report. See the rest ↓
Sample pages from an earlier edition. The report has been updated with new data since.
What metrics actually tell us who wins quantum computing?
The metric that matters is not “qubit count” but “how many reliable logical operations can the machine execute before the computation dies.”
When we dug through the recent announcements, the useful pattern is clear: raw hardware quality matters, but it only becomes strategically decisive when it converts into logical qubits, logical gates, repeated correction cycles, and circuits that cannot be reproduced classically.
The best way to read the race is therefore in layers.
First, we check whether the physical machine is good enough: two-qubit fidelity, measurement fidelity, connectivity, crosstalk, coherence, and circuit depth.
Then we check whether that quality survives error correction: logical error rate, break-even versus physical qubits, number of logical qubits, logical two-qubit gates, and real-time decoding.
Finally, we check whether the system produces something externally meaningful: verifiable quantum advantage, customer access, repeatable benchmarks, and useful application-level circuits.
| Metric | Why it matters |
|---|---|
| Logical error rate per cycle / per gate | This is the closest metric to “can the computer run a long useful computation?” Physical fidelity is upstream; logical error rate is the thing fault-tolerant computing ultimately needs. It is also where the race moved most in 2026, with Quantinuum publishing new logical-fidelity records and Google a new magic-state result. |
| Break-even logical performance | A logical qubit only matters if it beats the best unencoded physical version. Google showed below-threshold scaling on Willow; Quantinuum has shown encoded computations beating unencoded ones. |
| Number of useful logical qubits | One beautiful logical qubit is a physics milestone; dozens of logical qubits start to look like a computing architecture. This is why Quantinuum’s 48 error-corrected / 94 error-detected logical-qubit claims matter. |
| Logical two-qubit gates | Single logical qubits are not enough. A fault-tolerant machine needs entangling operations between logical qubits, otherwise it cannot run general algorithms. |
| Repeated QEC cycles with real-time decoding | Error correction is only scalable if decoding happens fast enough while the processor is running. Google reported real-time decoding on Willow; IBM and Quantinuum have made real-time decoding and control central to their roadmaps. IonQ’s September 2026 decoder result ran on simulated circuits, not on a quantum processor. |
| Two-qubit gate fidelity | This is still the single most important physical-level metric because two-qubit gates are usually the noisiest operations. IonQ’s >99.99% claim is therefore technically important, but it is not the same as proving logical execution. |
| Circuit volume: width × depth × fidelity | A useful processor must run wide and deep circuits without collapsing into noise. IBM’s emphasis on 5,000+ two-qubit-gate workloads and Quantinuum’s RCS/fidelity results are more informative than qubit count alone. |
| Independent verifiability of advantage claims | Quantum advantage is only persuasive if the task is checkable or classically benchmarked hard. Google’s Quantum Echoes is strong because it was framed as a verifiable quantum-advantage algorithm, not just random sampling. |
| Cloud/customer access | This tells us whether the machine is a lab artifact or an operating platform. IBM, Quantinuum, and IonQ are more open here than Google; Google’s Willow access is still selective, not broad public cloud access. |
| Roadmap credibility: dated, quantified, architecture-backed targets | Roadmaps are weak when they are just qubit counts. They become useful when they include logical qubits, gate counts, architecture choices, decoding, manufacturing, and dates. IBM is strongest here with 200 logical qubits and 100M gates by 2029; Quantinuum’s Apollo, also due in 2029, is less quantified in public. |
Prefer verified numbers to headlines? Our report is here ↓
Sample pages from an earlier edition. The report has been updated with new data since.
Are quantum companies transparent on the metrics that matter?
No, not fully. The industry is transparent enough to compare directionally, but not transparent enough to score everyone on the same technical sheet.
The biggest problem is that each company chooses the metric where it looks strongest: Google publishes deep scientific milestones, Quantinuum publishes concrete logical and error-correction evidence, IBM publishes roadmap and ecosystem metrics, and IonQ publishes physical-fidelity claims and, as a listed company, fast-growing revenue.
The clean read is: Quantinuum and Google are the most technically transparent; IBM is the most roadmap-transparent; IonQ is the most transparent on sales and the most selective on logical computing.
Two things improved the picture in 2026. Quantinuum now files quarterly results with the SEC, so its revenue, cash and roadmap dates are on the record. And DARPA’s Stage C will put IBM’s and IonQ’s hardware in front of an independent verification team through 2029. Neither fixes the core gap: there is still no common, audited logical-qubit benchmark across the four.
That does not mean IonQ’s metrics are fake. It means its public evidence is thinner on the layer that decides fault-tolerant leadership: multi-logical-qubit execution on real hardware. Its September 2026 real-time decoder covered up to 408 logical qubits, but in simulated circuits.
| Company | Transparency score | Why |
|---|---|---|
| Quantinuum | 90 / 100 | Best transparency on the metrics that matter most for fault tolerance: gate fidelities, all-to-all architecture, logical-qubit demonstrations, logical-fidelity records, papers and preprints, and now SEC-filed financials and dated Sol (2027) and Apollo (2029) targets. The caveat: some logical-qubit work relies on error detection, postselection and code choices, and the October 2026 record is detailed in a preprint, not yet peer-reviewed. |
| 84 / 100 | Very strong scientific transparency: Nature-level Willow QEC data, below-threshold scaling, distance-7 surface code, RCS benchmark, Quantum Echoes, and real-time decoder details. The weak point is access and operating transparency: Willow is not broadly available to public customers, and Google is less explicit than IBM on a dated commercial FTQC system roadmap. | |
| IBM | 78 / 100 | Best roadmap transparency: named systems, 2029 Starling target, 200 logical qubits, 100M gates, Nighthawk/Loon architecture, 300mm fabrication, a $10B five-year investment and, since October 2026, DARPA Stage C testing. Its July 2026 run on 70 error-detected logical qubits is public, with circuits posted online, but it is a preprint and has less logical-error data than Google or Quantinuum publish. |
| IonQ | 65 / 100 | Strong on headline physical metrics and on financials: >99.99% two-qubit fidelity, #AQ 64 and quarterly revenue are public and specific. But IonQ is less transparent on comparable logical-computing evidence: fewer hardware demonstrations of repeated QEC cycles, multi-logical-qubit algorithms or logical gates than Quantinuum or Google. It says it plans to publish error-correction results on its own hardware. |
If you want more recent data on this point, see the latest figures in our Quantum Computing report ↓.
Want to know who is really ahead? Our report is here ↓
Sample pages from an earlier edition. The report has been updated with new data since.
What are the most impressive quantum metrics published recently?
The most impressive recent quantum metrics are not the biggest numbers but the ones that move from physics demo toward fault-tolerant computation.
Over the last 18 months, Quantinuum and Google published the strongest “proof” metrics, and in 2026 the contest moved squarely to logical fidelity. IBM published the strongest industrialization metrics, and IonQ the cleanest physical-fidelity headline.
The ranking below is intentionally strict. A metric ranks higher when it is closer to useful fault-tolerant computation, externally interpretable, difficult to fake with a narrow benchmark, and connected to a published paper or concrete system.
| Rank | Company and metric | Why this ranking | When |
|---|---|---|---|
| 1 | Quantinuum: 99.9995% logical memory fidelity on Helios with a distance-7 color code, about five logical errors per million error-correction rounds | This is now the strongest logical-error metric in the set. Quantinuum says it is 10× better than its own September 2026 Helix-code record and about 100× better than Google’s best published result. It ranks first because logical error rate is the metric fault tolerance depends on; the caveats are that it is company-reported, measures stored memory rather than a full computation, and is detailed in a preprint, not yet a peer-reviewed paper. | Oct. 2026 |
| 2 | Google: Quantum Echoes ran 13,000× faster than the best classical algorithm on a top supercomputer | This is the strongest algorithmic milestone because Google framed it as a verifiable quantum-advantage task, not just a random benchmark. The output can be checked against physics structure, which makes the advantage claim more meaningful. | Oct. 2025 |
| 3 | Quantinuum: 48 error-corrected logical qubits and 94 error-detected logical qubits on 98 physical qubits | This is the strongest logical-qubit scaling metric. The codes and postselection details require interpretation, but the conversion of hardware into many logical qubits is exceptional. | Mar. 2026 |
| 4 | Google: below-threshold surface-code QEC on Willow, distance-7 with 0.143% error per cycle | One of the cleanest QEC results in the field: increasing code distance reduced logical error, which is exactly the direction fault tolerance requires. | Dec. 2024 / Nature 2025 |
| 5 | Google: magic state cultivation on Willow, with 99.99% T-state fidelity and about 40× lower error than uncultivated states | Published in Science, this attacks one of the costliest parts of a future fault-tolerant machine: preparing the magic states that universal algorithms need. It ranks below the surface-code result because it is one component, not a full logical computation. | Oct. 2026 |
| 6 | IBM + University of Chicago: 70 error-detected logical qubits, 2,415 logical two-qubit gates and 468 logical T gates in a hard sampling task, run in about 15 minutes | IBM’s strongest logical-computing result so far, with logical errors about 10× lower than physical ones and a statistical bound on fidelity that IBM calls verifiable quantum advantage. It ranks here because it relies on error detection and discarding flagged runs, and it is a preprint. | Jul. 2026 |
| 7 | Google: Willow RCS benchmark in under five minutes vs 10 septillion years classical estimate | The number is spectacular and useful as a system-level stress test. It ranks lower because RCS is not directly useful and has a history of classical-simulation debates. | Dec. 2024 |
| 8 | Quantinuum Helios: 98-qubit trapped-ion processor with all-to-all connectivity and 7.9×10⁻⁴ average two-qubit infidelity | A very strong full-system hardware metric because it combines scale, fidelity and architecture, averaged across operational zones rather than a single best pair. | Nov. 2025 |
| 9 | Quantinuum: fault-tolerant T-gate demonstration on Helios | A T gate is a building block for universal fault-tolerant computation. It ranks below the logical-fidelity record because it is a single gate type, not a full algorithm. | 2026 |
| 10 | IonQ: >99.99% two-qubit gate fidelity without ground-state cooling | Still the best pure physical-fidelity metric in the set. It ranks below the logical-QEC metrics because high physical fidelity is necessary but not sufficient to prove logical computation. | Oct. 2025 |
| 11 | IBM: Starling target of 200 logical qubits and 100M gates by 2029 | The most concrete industrial FTQC target among the four. It specifies logical qubits and gate count, not just physical qubits, but it is still a roadmap rather than a demonstrated machine. | Jun. 2025 |
| 12 | IBM and IonQ: selected for Stage C of DARPA’s Quantum Benchmarking Initiative | Stage C means DARPA’s independent team will test their hardware through 2029; IonQ’s agreement has a ceiling of up to $300M, not a guaranteed amount. It ranks here because it is outside validation of their plans, not a performance result. Quantinuum, in Stage B since November 2025, was not in this batch. | Oct. 2026 |
| 13 | Quantinuum + Microsoft: 12 highly reliable logical qubits | An important bridge from “a few logical qubits” to a multi-logical-qubit register, since overtaken by Helios results. | Sep. 2024 |
| 14 | Quantinuum + Microsoft: 800× better logical error rate than physical error rate | A clean break-even-style metric: error correction improved the computation rather than adding noise. Older, but still foundational for judging Quantinuum’s lead. | Apr. 2024 |
| 15 | Google: real-time decoding on Willow with 63 µs average latency at distance-5 | Fault tolerance is a live control loop, not just a code. It ranks below the headline QEC scaling because it is an enabling metric. | Dec. 2024 |
| 16 | Google AlphaQubit 2: real-time decoding faster than 1 µs per cycle up to distance 11 on commercial accelerators | A strong scalability signal for the classical side of QEC, but decoder infrastructure rather than a hardware demonstration. | Dec. 2025 |
| 17 | IonQ: end-to-end real-time QEC decoder on a single standard CPU, tested on circuits simulating up to 408 logical qubits and 31.5M operations | Useful evidence that decoding need not be a bottleneck for IonQ’s architecture. It ranks lower because the circuits were simulated, not run on an IonQ processor, and no logical error rates were reported. | Sep. 2026 |
| 18 | IBM: $10B five-year quantum commitment | Not a physics metric, but leadership now requires fabrication, packaging, software, partnerships and M&A. Money alone does not prove performance. | Jun. 2026 |
| 19 | IBM: Loon chip demonstrates hardware elements for fault-tolerant quantum computing | A credible architecture milestone, less directly comparable to a demonstrated logical-qubit benchmark. | Nov. 2025 |
| 20 | IBM: Nighthawk with 120 qubits and 218 tunable couplers | IBM is improving connectivity and workload complexity, not just adding qubits. Still physical-layer progress. | Nov. 2025 / Jan. 2026 access |
| 21 | IBM: Nighthawk designed for 5,000 two-qubit-gate workloads | A better IBM metric than qubit count because it speaks to circuit depth, but deep noisy circuits are not fault-tolerant circuits. | Nov. 2025 |
| 22 | Quantinuum Helios: 99.9975% single-qubit fidelity and 99.921% two-qubit fidelity reported for commercial system | Commercial-system fidelity, not a lab-isolated headline. It ranks below IonQ’s 99.99% on peak fidelity but is tied to a launched system. | Nov. 2025 |
| 23 | IonQ: #AQ 64 on its 100-qubit Tempo system | A strong application-style benchmark and commercial signal, but #AQ is partly IonQ’s own framing and less universally interpretable than logical error rate. | Sep. 2025 |
| 24 | Quantinuum: Helios real-time control engine for dynamic programs | Practical QEC needs dynamic circuits and low-latency control. Architectural unless paired with specific logical-circuit metrics. | Nov. 2025 |
| 25 | Quantinuum + NVIDIA: real-time QEC decoding with NVQLink / CUDA-Q integration | The decoding bottleneck is moving into systems engineering. An infrastructure proof point rather than the end output. | Nov. 2025 / Mar. 2026 |
| 26 | IBM: QEC decoding 10× speedup over current leading approach | IBM’s 2029 roadmap depends on decoding speed. Less directly comparable than Google’s published real-time latency. | Nov. 2025 |
| 27 | IBM: move to 300mm wafer fabrication to double development speed and raise chip complexity 10× | One of IBM’s strongest industrialization metrics. Manufacturing speed does not prove quantum performance, but it matters for scaling. | Nov. 2025 |
| 28 | Quantinuum: 56-qubit H2-1 RCS result with 100× improvement over prior Google 2019 benchmark | A strong trapped-ion system-level benchmark, now outside the 18-month window but still useful context. | Jun. 2024 |
| 29 | Google: Willow Early Access Program for selected external researchers | Google moved from closed demonstrations toward external scrutiny. Selective access, not broad cloud availability. | Mar.–Apr. 2026 |
| 30 | IonQ: roadmap of 1,600 logical qubits in 2028 and 80,000 by 2030 | Ambitious and commercially important, but a target. IonQ’s public logical-computing proof points on hardware lag the claim. | 2025–2026 |
If you want more recent data on this point, see the latest figures in our Quantum Computing report ↓.
We made a report out of all this. Take a look ↓
Sample pages from an earlier edition. The report has been updated with new data since.
Has Quantinuum overtaken Google in quantum computing?
Yes, on the metric that matters most for fault tolerance, and narrowly overall. No, on verifiable quantum advantage, where Google still holds the strongest result.
The shift comes from logical fidelity. In October 2026, Quantinuum reported 99.9995% logical fidelity on Helios, after a September 2026 record with its Helix code that used only 10 physical qubits per logical qubit. Google answered the same month with a Science paper on magic state cultivation, reaching 99.99% fidelity on T-states. Both are real progress. The two figures measure different operations, so they are not a clean head-to-head. Quantinuum’s is higher, but it comes from a company preprint, while Google’s is peer-reviewed.
The tension is that Google still has the only peer-reviewed verifiable advantage result of the four (IBM’s July 2026 claim is a preprint), and its superconducting chips run cycles far faster than trapped ions. Speed will matter once machines run millions of logical operations. For now, accuracy per logical operation is the scarcer resource, and Quantinuum has more of it in public.
Our aggregate ranking would be:
| Rank | Company | Why |
|---|---|---|
| 1 | Quantinuum | Quantinuum is closest to turning hardware quality into logical computing. Its 99.9995% logical fidelity, 48 error-corrected and 94 error-detected logical qubits, fault-tolerant T gate and Helios system fidelity make it the strongest execution story. Its cash of about $2.1 billion at the end of June 2026, after the IPO, is meant to fund Sol in 2027 and Apollo in 2029. |
| 2 | Google still has two of the highest-impact proof metrics: Willow’s below-threshold surface-code QEC and Quantum Echoes’ 13,000× verifiable quantum-advantage claim, now joined by peer-reviewed magic state cultivation. It falls to second because its public logical-fidelity results trail Quantinuum’s and Willow access stays selective. | |
| 3 | IBM | IBM’s metrics are less about “look what we already proved” and more about “look how we will industrialize it.” The 2029 Starling target, 200 logical qubits, 100M gates, Nighthawk/Loon, 300mm fabrication, the $10B commitment and DARPA Stage C make IBM the most credible platform-scale roadmap. Its July 2026 run on 70 error-detected logical qubits adds proof, but not yet peer-reviewed proof. |
| 4 | IonQ | IonQ has the biggest business, with $80.1M of revenue in the second quarter of 2026, and one of the best physical metrics: >99.99% two-qubit fidelity. But if we score demonstrated progress toward logical computing on real hardware, IonQ has fewer public proof points than Quantinuum or Google. |
If you want more recent data on this point, see the latest figures in our Quantum Computing report ↓.
Is IonQ catching up in quantum computing, or just selling more?
IonQ is selling much more, and catching up only partly. Its business lead is real; its technical lead on logical computing is not shown yet.
The numbers are large. IonQ reported $80.1 million of revenue in the second quarter of 2026, up 287% from a year earlier, and raised its 2026 guidance to $280 million to $290 million, excluding SkyWater, the foundry it bought for about $1.8 billion and took over in July 2026. Quantinuum, by comparison, guides $28 million to $32 million for the year.
The technical record is thinner. IonQ’s headline is still the >99.99% two-qubit fidelity reported in 2025. A best-in-class physical gate does not prove repeated error correction, logical gates or useful logical circuits. Its September 2026 real-time decoder is a solid engineering result, but it ran on simulated circuits. Its 256-qubit Superion system has a first buyer, Florida International University, with installation expected in late 2027. IonQ says error-correction results on its own hardware are coming.
The DARPA Stage C selection is IonQ’s strongest outside signal: independent testers will check multiple Superion generations through 2029. But the agreement’s “up to $300 million” is a ceiling, and funding beyond the first tranche depends on future appropriations.
We would also be careful with IBM roadmap metrics. Targets like “200 logical qubits and 100M gates by 2029” are not demonstrated capability, and Stage C starts a test rather than ends one. IBM deserves credit for specificity, and its July 2026 run on 70 logical qubits with 2,415 logical two-qubit gates is a real step. But it relied on error detection, discarding flagged runs, not full error correction, and it is not yet peer-reviewed. It should not be scored as ahead in logical execution yet.
Google has the opposite problem: its metrics are scientifically strong, but some headlines sound more commercially mature than they are. The Willow RCS “10 septillion years” number is not a useful application, and Willow remains limited to selected researchers.
If you want more recent data on this point, see the latest figures in our Quantum Computing report ↓.
If this is useful, the full report goes much further. Have a look ↓
Sample pages from an earlier edition. The report has been updated with new data since.
Our methodology
This analysis tests who is currently ahead in quantum computing among Quantinuum, IonQ, IBM, and Google, based on the evidence available today. We compare the companies across demonstrated technical results, operating-platform maturity, and dated roadmap credibility.
The 23-criteria table is a breadth check, not a weighted winner model. It shows where public evidence exists, while the final ranking gives more weight to proof points that move closer to fault-tolerant computing.
For recent metrics, we focused mainly on the last 18 months, while keeping a small number of 2024 results where they explain a company’s current position. We ranked mixed signals, benchmarks, logical-qubit results, fidelity claims, access programs, investment commitments, and roadmap targets, by how directly they show progress toward useful, reliable quantum computation.
We used company-specific metrics when they were concrete and sourceable, but weighted them more carefully when they were less standardized across the industry. “Unclear” means the public evidence was not specific enough to score confidently; it does not mean the company has not done the work.
We gave the most weight to metrics that move from physical hardware toward logical computation: logical error rates, break-even logical performance, repeated QEC cycles, real-time decoding, logical two-qubit gates, verifiable quantum advantage, and application-level circuits.
We treated roadmap claims differently from demonstrated results. IBM’s Starling target, Quantinuum’s Apollo target, IonQ’s future logical-qubit ambitions, and DARPA contract ceilings matter, but they are not scored the same way as published logical-qubit, QEC, fidelity, or advantage demonstrations.
We also separated customer-access evidence from technical leadership evidence. IBM, Quantinuum, and IonQ look stronger on platform availability, while Google looks stronger on scientific proof but weaker on broad public access to Willow.
Key sources used for this analysis include: Quantinuum on its 99.9995% logical-fidelity record, Quantinuum’s second-quarter 2026 results filed with the SEC, IonQ’s second-quarter 2026 results, IonQ on its real-time quantum error decoder, IBM on advancing to Stage C of DARPA’s Quantum Benchmarking Initiative, the University of Chicago on its 70-logical-qubit experiment with IBM, Quantum Computing Report on Google’s magic state cultivation on Willow, Nature on Google’s below-threshold surface-code QEC, and Google on Quantum Echoes and verifiable quantum advantage.
Additional key sources include: Quantinuum’s Helios system page, Quantinuum’s Helios launch note, arXiv on Quantinuum logical-qubit work, Quantinuum on skinny logical codes, arXiv on later Quantinuum logical-code work, Microsoft Azure Quantum on 12 logical qubits with Quantinuum, Microsoft on Quantinuum logical qubits with 800× better error rate, arXiv on Quantinuum/Microsoft logical-qubit results, Quantinuum’s 56-qubit trapped-ion announcement, and arXiv on Quantinuum’s 56-qubit benchmark work.
Other key sources include: IonQ on >99.99% two-qubit gate fidelity, arXiv on IonQ’s high-fidelity trapped-ion work, IonQ on #AQ 64, IBM on its large-scale fault-tolerant quantum computing roadmap, and IBM’s $10B quantum commitment announcement.
Written by the New Market Pitch teamWe track new markets so founders and investors can move faster. About us
