DARPA (US Defense Advanced Research Projects Agency) · 2026-10-07
Ranks: companies
DARPA's QBI tests whether a company's plan for a useful, fault-tolerant quantum computer could be built and run as intended by 2033. Companies pass through Stage A (a concept), Stage B (a year-long review of R&D plans and risk-reduction prototypes) and Stage C, where a government evaluation team tests whether the proposed utility-scale system works as designed. Selections are announced on staggered timelines, and DARPA expects more teams to advance.
- Atom Computing, entered Stage C (neutral atoms)
- Diraq, entered Stage C (silicon CMOS spin qubits)
- IBM, entered Stage C (modular superconducting)
- IonQ, entered Stage C (trapped ions)
- Microsoft and PsiQuantum, already in the equivalent final phase via the US2QC pilot (6 teams in total)
- Who's behind it
- Run by a US government agency with a government verification team. Participants are vendors in a government programme, not submitters to a public leaderboard. DARPA says QBI is not meant to pick a few winners, and that each approach is judged on its own merits.
- How much weight to give it
- An authoritative government selection, but it assesses plans and their feasibility, not measured performance, and the October 2026 announcement says nothing about the other seven Stage B companies.
Free · Read it at DARPA (US Defense Advanced Research Projects Agency) →
Jülich Supercomputing Centre authors (J. A. Montanez-Barrera, Kristel Michielsen); arXiv preprint · 2026-10-05
Ranks: machines
The authors built a low-shot test of the building blocks that error correction needs: parity extraction, mid-circuit measurement, reset, feed-forward and scheduling. They ran it on 10 commercial processors from Quantinuum, IBM and IQM, including circuits shaped like surface codes, colour codes and qLDPC codes. Signal decay is converted into an effective hardware error rate so the machines can be compared. On ibm_phoenix the score correlated with a real logical-memory experiment.
- Quantinuum (Helios-1, H2-1), best overall across tested QEC primitives (effective hardware error), λeff about 0.22% to 0.49% (IBM about 1.1% to 3.2%)
- Quantinuum Helios, hardware error vs top superconducting processors tested, about 13x lower at 30 data qubits; about 8x lower at 50 data qubits
- Quantinuum, 91-data-qubit triangular colour code signal, r = 0.733
- Who's behind it
- Academic (Jülich) authors and no vendor co-authors listed on the abstract page. Funding and hardware access are not described in the pages opened. Quantum Computing Report's article links to a Quantinuum blog post about the result.
- How much weight to give it
- Very recent preprint, not yet peer reviewed; the per-device numbers were read from a trade-press summary, not from the paper itself.
Free · Read it at Jülich Supercomputing Centre authors (J. A. Montanez-Barrera, Kristel Michielsen); arXiv preprint →
MIT Initiative on the Digital Economy (with Accenture) · 2026-10
Ranks: machines
MIT compiles a dataset of about 300 quantum processors (235 unique products, 56 commercially available from 24 manufacturers) from manufacturer announcements, public datasets, web searches and direct queries to providers. It charts physical metrics (qubit count, gate fidelities, coherence, gate speed), aggregate metrics (Quantum Volume, logical qubits) and the rare application benchmarks, by device and by modality. The report is annual. It does not run any tests itself.
- Trapped ions (modality best), 2-qubit gate fidelity, 99.92%
- IBM Heron r2 – Fez, 2-qubit gate fidelity, single device shown on chart, 99.70%
- IQM Garnet, fastest published 2-qubit gate, 48 ns
- Rigetti Cepheus-1-108Q, 2-qubit gate speed (second), 60 ns
- IBM Condor (2023), qubit count, superconducting, 1,121
- Who's behind it
- Academic team at MIT IDE in collaboration with Accenture, supported by an NSF Engineering Research Centers grant. The data is mostly self-reported by vendors, and the report itself warns that vendors highlight favourable metrics under differing definitions.
- How much weight to give it
- Broad and well-sourced overview, but it compiles vendor-reported numbers rather than measuring them, and it says no independently verified universal metric exists.
Free · Read it at MIT Initiative on the Digital Economy (with Accenture) →
Unitary Foundation · 2026-03-09
Ranks: machines
Metriq runs the same suite of tests on cloud-accessible quantum computers through its own open-source runner, metriq-gym. The suite mixes system-level tests (entanglement connectivity, error per layered gate, mirror circuits, circuit speed) with application-style tests such as QAOA, QFT and a quantum machine learning kernel. Each test is scaled against a baseline machine set to 100 (currently IBM's ibm_torino), and the results are combined into one composite Metriq Score that gives more weight to wider circuits. Results live in a public, versioned GitHub dataset and are added through reviewed pull requests. No fixed update schedule is stated.
- ibm_boston, Metriq Score (composite), 252.61
- quantinuum_h2_2, Metriq Score (composite), 188.05
- ibm_pittsburgh, Metriq Score (composite), 174.51
- ibm_kingston, Metriq Score (composite), 174.23
- ibm_torino, baseline device (score fixed at 100), 100.00
- Who's behind it
- Run by the non-profit Unitary Foundation. The maintainers set the test parameters, and outside operators may submit results, which are merged after review. Per the paper's acknowledgments, funding includes the US Department of Energy and an IBM Sponsored Research Agreement; IBM benchmarks ran on an IBM Premium account; OriginQ supplied runtime; Quantinuum access came through Oak Ridge's user programme. IBM machines lead the score and IBM is a funder, which is a possible bias even though the method is open.
- How much weight to give it
- Open, reproducible method run by a non-profit, but it covers only about a dozen cloud machines, a missing benchmark scores zero, the composite depends on the chosen baseline, and IBM both funds the work and tops the table.
Free · Read it at Unitary Foundation →
IEEE Access (authors: M. Vukšić, J. Ćelić, A. Cuculić, University of Rijeka) · 2026-03-05
Ranks: platforms
This is a peer-reviewed literature review that compares qubit platforms (superconducting, trapped ion, photonic, neutral atom, silicon spin, topological) and named processor families. It uses published figures for qubit count, gate and readout fidelity, coherence, speed, Quantum Volume and Algorithmic Qubits, and it publishes a comparison table (Table 1). It is a single publication with no updates.
- Quantinuum H1-1, Quantum Volume, described as a record, QV 8192 (dated 2021 in the review)
- IonQ Forte, Algorithmic Qubits (vendor metric), #AQ 29
- IBM Heron, superconducting gate fidelity as quoted, >99.9% single-qubit, 99.5% two-qubit
- Who's behind it
- Academic authors funded by a University of Rijeka grant (for an unrelated maritime project, per the funding note); no vendor authors. Figures are compiled from vendor and lab publications, and the authors note they reflect best-case calibrated results.
- How much weight to give it
- Peer-reviewed, but the paper itself says Table 1 is 'a high-level comparative snapshot rather than a definitive ranking'. Its data is already dated, and some milestone dates should be checked against primary sources.
Free · Read it at IEEE Access (authors: M. Vukšić, J. Ćelić, A. Cuculić, University of Rijeka) →
Community effort launched by IBM with Algorithmiq, Flatiron Institute and BlueQubit (15+ contributing organisations) · 2026-02-23
Ranks: approaches
The tracker is an open log of claimed quantum-advantage experiments in three pathways: observable estimation, variational problems and classically verifiable problems. Each candidate is listed alongside the best classical methods, and the community tries to match or beat it classically. Entries are marked active, superseded or baseline. It is updated continuously as results and challenges arrive.
- peaked_circuit / P12_Hqap_98x2457, classically verifiable problems (active)
- peaked_circuit / P11_Hqap_98x1999, classically verifiable problems (active)
- floquet_mixed_field_ising / zzd3_51qx16c, observable estimation (active)
- su2_hadron_dynamics_lsh / x_100_meson, observable estimation (active)
- Who's behind it
- Initiated and promoted by IBM, a vendor. Per IBM's blog, its submissions ran on IBM and Quantinuum hardware. Contributors also include universities and national labs (Caltech, Los Alamos, ORNL, RIKEN and others), NVIDIA and several start-ups. Likely to over-represent IBM hardware.
- How much weight to give it
- Useful for showing which advantage claims survive classical challenge, but it is vendor-initiated, does not rank machines, and IBM's own blog reports a featured candidate later beaten by new classical methods.
Free · Read it at Community effort launched by IBM with Algorithmiq, Flatiron Institute and BlueQubit (15+ contributing organisations) →
DARPA (US Defense Advanced Research Projects Agency) · 2025-11-06
Ranks: companies
In this list, DARPA names the 11 companies that moved from Stage A (concept descriptions) into Stage B, a year-long review of each company's detailed R&D plan, risks and prototypes toward a utility-scale quantum computer by 2033. Companies that succeed are invited to Stage C for government verification. The page also records that Google Quantum AI joined Stage A in September 2025.
- Atom Computing, Diraq, IBM, IonQ, Nord Quantique, Photonic Inc., Quantinuum, Quantum Motion, QuEra Computing, Silicon Quantum Computing, Xanadu, selected for Stage B (11 companies, unranked)
- Who's behind it
- US government programme with a government verification and validation (IV&V) process. Companies are participants rather than paying submitters. DARPA says multiple, a single, or no participants may show a viable path.
- How much weight to give it
- An official list of who advanced, not a ranking; it partly rests on plans and roadmaps, and Stage C has since narrowed it (see darpa-qbi-stage-c-2026).
Free · Read it at DARPA (US Defense Advanced Research Projects Agency) →
Jülich Supercomputing Centre (Forschungszentrum Jülich) and Purdue University; arXiv preprint · 2025-05-28
Ranks: machines
Academic researchers ran one fixed optimisation workload, linear-ramp QAOA on weighted MaxCut, on 24 cloud quantum processors from six vendors. They measured how long each machine keeps a usable signal as the circuit gets wider (up to 156 qubits) and deeper (up to 10,000 layers). Three problem layouts were used: a 1D chain, the chip's native layout, and a fully connected graph. This is a one-off study, not a continuously updated leaderboard, though Metriq has adopted the same test.
- quantinuum_H2-1, fully connected graphs: only platform to pass at 50 and 56 qubits (largest certified run), r = 0.872 at 56 qubits, p = 3
- ibm_fez, 1D chain (100 qubits) and native layout, r = 0.808 (1D chain, fractional gates); r = 0.771 (native layout, p = 10)
- ibm_fez, deepest run attempted, p = 10,000 (about 990,000 ZZ gates), signal coherent to about p = 300
- ionq_forte_enterprise, fully connected graph pass at 28 qubits, r = 0.770
- Who's behind it
- Academic authors. Funding came from the German BMBF and a Purdue startup grant. IBM access was via the Quantum Collaborative (the paper states IBM does not endorse its views), Quantinuum access came through Oak Ridge with a capped budget, and some IBM results were community-contributed. Vendors did not author the paper.
- How much weight to give it
- Careful, independent academic study using one workload; it is a 2025 snapshot that predates newer machines, and the access budgets differed by vendor.
Free · Read it at Jülich Supercomputing Centre (Forschungszentrum Jülich) and Purdue University; arXiv preprint →
