Independent benchmark
Hardware Honesty Tracker
Every major quantum computer, side by side, with the numbers that actually matter. Sourced from primary calibration pages and roadmap announcements — no press releases, no vendor PDFs, no marketing claims without a source link.
- Systems tracked
- 13
- Vendors
- 10
- Last verified
- 2026-04-01
Why this page exists
Every hardware vendor benchmarks themselves favorably. None can publish side-by-side numbers showing their competitor wins on a metric. We can, because we don't sell quantum compute — we just teach. Independence is the entire point. More on our editorial stance →
Superconducting
4 systems| System | Qubits | 2q error | 1q error | Readout | T₁ | Connectivity | Access | Source |
|---|---|---|---|---|---|---|---|---|
| IBM Heron r2 released 2024 Q4 | 156 | 3×10^-3 | 2×10^-4 | 1.50% | 300 µs | Heavy-hex | Free tier | link ↗ |
| IBM Nighthawk released 2025 Q4 | 1500 | 2×10^-3 | 2×10^-4 | 1.00% | 350 µs | Heavy-hex | Paid cloud | link ↗ |
| Google Quantum AI Willow released 2024 Q4 | 105 | 4×10^-3 | 1×10^-3 | 1.00% | 100 µs | Square grid | Private only | link ↗ |
| Rigetti Cepheus-1 released 2026 Q1 | 108 | 5×10^-3 | 5×10^-4 | 2.00% | 40 µs | Square grid | Paid cloud | link ↗ |
Trapped-ion
4 systems| System | Qubits | 2q error | 1q error | Readout | T₁ | Connectivity | Access | Source |
|---|---|---|---|---|---|---|---|---|
| Quantinuum H2 released 2023 | 56 | 1×10^-3 | 3×10^-5 | 0.30% | 50 s | All-to-all | Paid cloud | link ↗ |
| Quantinuum Helios released 2025 Q4 | 96 | 8×10^-4 | 2×10^-5 | 0.20% | 60 s | All-to-all | Paid cloud | link ↗ |
| IonQ Forte Enterprise released 2024 | 36 | 4×10^-4 | 2×10^-4 | 0.50% | 60 s | All-to-all | Paid cloud | link ↗ |
| IonQ Tempo released 2025 Q4 | 64 | 1×10^-4 | 3×10^-5 | 0.30% | 60 s | All-to-all | Paid cloud | link ↗ |
Neutral-atom
3 systems| System | Qubits | 2q error | 1q error | Readout | T₁ | Connectivity | Access | Source |
|---|---|---|---|---|---|---|---|---|
| QuEra Aquila / Gemini-class released 2026 Q1 | 448 | 5×10^-3 | 1×10^-3 | 1.00% | 20 s | Programmable | Paid cloud | link ↗ |
| Pasqal Orion released 2025 | 1024 | 7×10^-3 | 2×10^-3 | 1.50% | 5 s | Programmable | Paid cloud | link ↗ |
| Atom Computing Phoenix released 2024 | 1180 | 4×10^-3 | 1×10^-3 | 0.80% | 40 s | Programmable | Private only | link ↗ |
Photonic
1 system| System | Qubits | 2q error | 1q error | Readout | T₁ | Connectivity | Access | Source |
|---|---|---|---|---|---|---|---|---|
| Xanadu Borealis (photonic) released 2022 | 216 | — | — | — | photon loss dominates | Measurement-based | Paid cloud | link ↗ |
Annealing
1 system| System | Qubits | 2q error | 1q error | Readout | T₁ | Connectivity | Access | Source |
|---|---|---|---|---|---|---|---|---|
| D-Wave Advantage2 released 2025 | 4400 | — | — | — | annealer — gate-error model doesn't apply | Square grid | Paid cloud | link ↗ |
Methodology
Every row links to a primary source — a vendor's official calibration page, an arXiv paper with verifiable numbers, or a roadmap announcement from a primary press release. We do not include numbers from analyst reports, slide decks, or anything we can't reproduce.
Where vendors publish ranges or moving averages, we record the most recently disclosed value. Trapped-ion and neutral-atom T₁ values are in seconds, not microseconds — formatted automatically.
Two-qubit gate error is the single most important number on this page. For a circuit with n two-qubit gates and per-gate error p, expected fidelity ≈ (1−p)n. A 100-CNOT circuit on a 10⁻³-error machine runs at ~90% fidelity; on a 5×10⁻³-error machine, ~60%. The column matters more than the qubit count.
Photonic and annealing systems use different error models — we mark their cells "—" rather than pretending the numbers compare directly.
Reading the data
- Lowest 2q error on commercial hardware right now: IonQ Tempo at 1×10⁻⁴, followed by Quantinuum Helios at 8×10⁻⁴. Both trapped-ion.
- Largest qubit count on commercial hardware: D-Wave Advantage2 at 4,400 (annealer), Atom Computing Phoenix at 1,180 (gate-model), Pasqal Orion at 1,024.
- Best coherence × gate-fidelity tradeoff: Quantinuum Helios. ~60s T₁, 8×10⁻⁴ 2q error, all-to-all connectivity.
- Best free-tier learning hardware: IBM Heron r2. 156 qubits, 10 minutes/month free, every tutorial works.
What this is and isn't
- Is: a snapshot of vendor-disclosed numbers, normalized for comparison.
- Isn't: an application-level benchmark. Real performance depends on circuit depth, problem structure, transpilation, error mitigation, and shot count — see tutorial 20 for the workflow.
- Isn't: a recommendation. Trapped-ion and superconducting solve different problems with different tradeoffs.
Want this updated faster?
The data is hand-curated today. The plan is daily-scraped from public vendor APIs by month 3. Subscribe for a one-line email when the tracker ships major updates: