State of Quantum Hardware — Week 28, 2026

Published 2026-07-12 · auto-generated from the public corpus · subscribe

This week's fleet

The Provenova corpus currently holds 28 device snapshots across six hardware providers: IBM (17 snapshots), IQM (3), Origin Quantum (3), Quantinuum (2), IonQ (2), and Rigetti (1). The most recent corpus refresh attempt ran on 2026-07-05. Rankings below are drawn from two source classes — vendor-reported specifications (manufacturer claims, not independently verified) and Metriq community benchmark submissions (CC-BY-4.0) — and the distinction matters when comparing across providers.

Two-qubit gate fidelity (vendor-reported). On the two-qubit fidelity board, Quantinuum's H2 leads at 0.999 (56 qubits, captured 2026-07-03), ahead of IonQ's Forte at 0.996 (36 qubits) and IQM's Garnet at 0.9951 (20 qubits). Rigetti's Ankaa-3 reports 0.995 across 84 qubits, though its figure was captured on 2024-12-23 and is the oldest data point in the top five. IonQ's Aria rounds out the group at 0.994 (25 qubits). The derived metrics attached to these entries fill out the picture: H2 carries a vendor-reported Quantum Volume of 33,554,432 (2^25), while Forte and Aria are described by their algorithmic-qubit counts of 36 and 25 respectively. Every entry on this board is vendor-reported — a manufacturer claim sourced from quantinuum.com, ionq.com, rigetti.com, and (for Garnet) the AWS Braket listing — so these should be read as best-case specifications rather than measured, redistributable results. The captured-at dates also vary by more than eighteen months across the top five, so a strict rank ordering conflates devices measured at very different points in their calibration history.

Error per layered gate (EPLG, Metriq). EPLG measures error accumulated per layer, so lower is better. On the EPLG board, IBM's ibm_boston leads at 0.00153921 (156 qubits, captured 2026-02-18), narrowly ahead of Quantinuum's h2-2 at 0.00179932 (56 qubits). IBM's ibm_pittsburgh (0.0027295) and ibm_torino (0.00550723, 133 qubits) follow, with IQM's iqm_garnet fifth at 0.00822524 (20 qubits). All EPLG figures come from Metriq submissions under CC-BY-4.0 and are redistributable.

Circuit layer operations per second (CLOPS, Metriq). The CLOPS board — a throughput measure where higher is better — is entirely IBM this week. ibm_fez tops it at 368,905, followed by ibm_kingston (362,372), ibm_boston (360,574), ibm_pittsburgh (358,613), ibm_marrakesh (353,642) and ibm_torino (345,670). All six are 156-qubit devices except ibm_torino (133 qubits), and all are Metriq-sourced.

QAOA approximation ratio (Metriq). On the QAOA approximation-ratio board (higher is better), Quantinuum's h2-2 leads at 0.8018 (56 qubits), ahead of IBM's ibm_boston (0.6734) and ibm_kingston (0.6412). IBM's ibm_marrakesh (0.5731) and ibm_pittsburgh (0.5695) fill the next two places, and IQM's iqm_garnet appears further down at 0.1432. The board's long tail is informative: IQM's iqm_emerald sits at 0.0099 (with a paired QFT-accuracy of 0.129) and Origin's wukong_72 records a negative ratio of -0.0350 (72 qubits, captured 2026-02-24), indicating its sampled solutions fell below the classical random baseline on that instance. Where a Metriq submission carries a QFT-accuracy value, Quantinuum's h2-2 is the strongest at 0.967, underlining that the same device tops both the fidelity-adjacent and optimisation-quality boards.

Movements

This is the first "State of Quantum Hardware" report published on Provenova, so there is no prior edition to compare against; the standings above describe the current boards as fetched. Future editions will track week-over-week movement from this baseline. A few structural points stand out in the current data. First, the boards separate cleanly by source class: the vendor-reported fidelity board is led by trapped-ion and other non-superconducting vendors (Quantinuum, IonQ, IQM), while the Metriq-benchmarked throughput board (CLOPS) is entirely superconducting IBM hardware. Second, Quantinuum's h2-2 is the only non-IBM device to place in the top two on both the EPLG and QAOA-ratio boards, appearing across both source classes. Third, IBM's hold on CLOPS reflects both the volume of IBM devices in the corpus (17 of 28 snapshots) and the availability of Metriq throughput submissions for its fleet — an absence of comparable submissions from other providers is not evidence of lower throughput, only of missing data.

New on the platform

Three research cards were published on the platform this week (ISO week 28), each pairing a small reference circuit with an arXiv attribution:

The platform's research-card corpus now holds 4 cards in total. Each card records the deterministic circuit hash for its reference execution.

Method note

Data on Provenova comes from three source classes, and each carries a different licence and confidence level. IBM calibration data is published under Apache-2.0; community benchmark results (EPLG, CLOPS, QAOA ratio, and related device metrics) are drawn from Metriq under CC-BY-4.0 and are redistributable; and vendor-reported specifications — the two-qubit fidelity board — are manufacturer claims, not independently verified, and are not redistributable as raw data. All Provenova platform runs, including the research-card reference circuits, are deterministic simulator executions rather than runs on physical hardware; the hardware metrics reported above are ingested from the sources named rather than measured by Provenova.