Compare the best publicly disclosed metrics horizontally by technology route. Values are taken from vendor announcements and peer-reviewed papers, with credibility marked by evidence levels E1–E5; updated weekly with the weekly journal.
| Field | SOTA anchor | Source / grade |
|---|---|---|
| Two-qubit gate fidelity (best of any platform) | 99.99%(Trapped ions: IonQ 2025-10 EQC prototype; Oxford Ionics electron-controlled gate error 8.4×10-5; SQC silicon-based donor 2025-12 also reported 99.99%) | E2–E3 |
| Highest all-pair F2Q in a commercial system | 99.921%(Quantinuum Helios, 98 qubits all-to-all, all pairs) | E2/E4 |
| Error suppression factor Λ public gold standard | Λ ≈ 2.1(Google Willow, surface code d=3→5→7) | E2 (Nature 2024) |
| Highest verified logical qubits | 96 LQ(QuEra, [[16,6,4]] code, 448 physical qubits, Nature 2026) | E2 |
| Best physical-to-logical encoding ratio | 2:1(Quantinuum, concatenated code, 48 LQ / 98 PQ); comparison: QuEra ≈4.7:1, Atom Computing ≈49:1, Google surface code 105:1 | E2 |
| Highest quantum volume | QV ≈ 33.5 M (≈225)(Quantinuum Helios) | E4 |
| Metric | SOTA reference | Representative system / source | Grade |
|---|---|---|---|
| F2Q (production system, median) | 99.5% – 99.9% (research devices >99.9%) | IBM Heron r2, IQM Radiance, Rigetti Cepheus-1(108q @99.5%) | E2–E4 |
| F1Q | ≥ 99.95% | cloud calibration data from leading vendors | E4 |
| t2Q | 20 – 100 ns | — | E2–E4 |
| F_RO | 98% – 99.5% | — | E4 |
| T1 | ~100 µs (Willow, ~5× improvement over Sycamore; leading devices 100–500 µs, note 2–3× TLS fluctuation) | Google Willow (Nature 2024) | E2 |
| N_ops = T2echo/t2Q | 103 – 104 | Template §4.2 | — |
| Physical qubits | 105 (Willow) / 120 (IBM Nighthawk, 218 tunable couplers, square lattice) / 1,121 (IBM Condor, no longer mainline) | — | E2–E4 |
| QEC | below-threshold surface code, Λ≈2.1, d=7; first demonstration of real-time error correction closed loop | Google Willow | E2 |
| Assessment note | Nighthawk takes the "fewer qubits–higher circuit complexity" route (circuit complexity +~30% over Heron); chips with >1000 qubits (Condor) have uncompetitive fidelity and must not receive bonus points for qubit count alone |
| Metric | SOTA reference | Representative system / source | Grade |
|---|---|---|---|
| F2Q | 99.99%(Single pair, IonQ EQC prototype 2025-10; Oxford Ionics "smooth gate" error 8.4(7)×10-5, no ground-state cooling required, n̄≤9.4 still ≲5×10-4);99.921%(Commercial all-pair, Helios) | arXiv 2510.17286;Quantinuum | E2–E4 |
| F1Q | 99.9975%(Helios) | — | E4 (with accompanying arXiv paper → E3) |
| t2Q | 10 – 500 µs | — | E2 |
| F_RO / SPAM | ≥ 99.9%;Helios SPAM 99.99% | — | E2–E4 |
| T2 | seconds–minutes (clock state) | — | E2 |
| N_ops | 103 – 105 | Template §4.2 | — |
| Physical qubits | 98 (Helios, Ba-137, QCCD + X junction + ring storage, all-to-all) | — | E2–E4 |
| QEC | 48 fully error-corrected LQs (2:1 concatenated code, single-shot correction, transversal logic) / 94 error-detection LQs with global entanglement / 50 LQ GHZ record; NVIDIA GB200 real-time decoding | Quantinuum | E2–E4 |
| QV | 33.5 M | Quantinuum | E4 |
| Assessment note | Platform common issue: qubit count in the tens to hundreds, repetition rate and ion transport overhead (IT-02) are scaling bottlenecks; IonQ AQ metric includes error mitigation, must be converted per template S-12 |
| Metric | SOTA reference | Representative system / source | Grade |
|---|---|---|---|
| F2Q (CZ) | 99.5% (60 atoms in parallel, Harvard/Lukin, Nature 2023); commercial: Atom Computing 99.6%, QuEra/Pasqal ~99.5% (including active error suppression) | Evered et al.; vendor disclosure | E2 / E4 |
| F1Q | ≥ 99.9% (production hardware) | — | E4 |
| t2Q | ~0.1 – 1 µs | — | E2 |
| F_RO | 97% – 99.5% | — | E2–E4 |
| T2 | Hyperfine ground state seconds; nuclear spin encoding ~40 s (Atom Computing); Rydberg gate related 1–10 ms | — | E2–E4 |
| N_ops | 103 – 104 | Template §4.2 | — |
| Array size | 1,180 physical qubits (Atom Computing, 1,225 sites); academic arrays thousands of atoms; Harvard 2025 demonstrated continuous operation (continuous loading) | — | E2–E4 |
| QEC | 96 LQ ([[16,6,4]] high-rate code, 448 PQ, all LQ simultaneous error correction gate operations, Nature 2026); 24 LQ entanglement (Atom Computing/Microsoft, Bacon-Shor, 1,180 PQ); toric code error reduction with scale demonstrated | QuEra;Atom Computing | E2 |
| Repetition rate | typical 1–10 Hz (S-09 hard constraint; continuous loading technique as improvement path) | — | E2–E3 |
| Assessment note | This route leads the entire industry in "verified logical qubit count" in 2026; weaknesses are mid-circuit measurement (NA-04) and cycle repetition rate; Google 2026-03 newly established neutral atom lab (Boulder/JILA) corroborates route momentum |
Per template §7.4, this route does not use F2Q metric, replaced by loss and success probability.
| Metric | SOTA reference | Representative system / source | Grade |
|---|---|---|---|
| SPAM (dual-rail encoding) | 99.98% ± 0.01% | PsiQuantum Omega(GlobalFoundries 300mm,Nature 2025) | E2 |
| HOM indistinguishability (independent sources) | 99.50% ± 0.25% | Same as above | E2 |
| two-qubit fusion fidelity | 99.22% ± 0.12% | Same as above | E2 |
| inter-chip interconnect fidelity | 99.72% ± 0.04% (42 m fiber) | Same as above | E2 |
| Detection efficiency | SNSPD ≥ 95% (due diligence threshold); leading devices higher | — | E2–E3 |
| System demo | Xanadu Aurora: 35 chips, 12 qubits, 13 km fiber-interconnected universal architecture; Borealis 216-mode GBS (sampling only, not universal) | — | E2 |
| Key materials | BTO (barium titanate) electro-optic switch 300mm process (PsiQuantum first) | — | E2 |
| Assessment note | All fidelities are conditioned on photon detection(Conditional fidelity)—fusion fidelity figures not accompanied by an end-to-end loss budget table (PH-01) must not be used for scoring; GBS sampling demonstrations must not be counted as universal computing capability (RF-8 / PH-05) |
| Metric | SOTA reference | Representative system / source | Grade |
|---|---|---|---|
| F2Q | 99.99%(SQC, atomically precise donor qubits, 2025-12, matching trapped-ion record);>99% on randomly sampled devices from 300 mm production wafers(Diraq+imec, Nature 2025-09, key result for manufacturing demonstration); 98.92% @ 1 K (Diraq hot qubits) | — | E2–E3 |
| F1Q | ≥ 99.9% (Diraq reported 99.85% @ 1 K) | — | E2–E3 |
| t2Q | ~10 – 100 ns | — | E2 |
| T2 | 10 – 100 µs (28Si purification) | — | E2 |
| N_ops | 102–103 (lowest among the five routes; evaluated jointly with gate timing per Template §4.2) | — | — |
| Scale | Device level ≤ ~12 qubits; Quantum Motion 1,024 quantum dot array characterized within 5 minutes (2025) | — | E2–E4 |
| QEC | 2026-01 first error detection;2026-03 first universal logical operation —— all "first" level, far behind other routes | — | E2–E3 |
| Commercial | Equal1 Bell-1 (GF 22FDX, F2Q 98.4%) delivered | — | E4 |
| Assessment note | selling point is mass production (SP-04):weight of cross-wafer/cross-batch distribution data should be higher than single-device peak; SQC's 99.99% is a single-device result from the donor route, cannot be directly extrapolated to the quantum dot mass production route |
https://qbitbrief.com/sota/ · Please credit the source when reprinting