[
  {
    "id": "quantum-volume",
    "name": "Quantum Volume",
    "short_name": "QV",
    "kind": "benchmark",
    "level": "system",
    "definition": "The largest square circuit, with as many layers as qubits, that a machine runs reliably; reported as 2 to the power of that size.",
    "unit": "quantum volume",
    "direction": "higher_is_better",
    "method_versions": [
      {
        "version": "Cross et al. 2019",
        "reference_url": "https://arxiv.org/abs/1811.12926"
      }
    ],
    "applies_to": [
      "gate-based machines"
    ],
    "comparability": "Comparable across gate-based machines when run to the published protocol. It grows exponentially, so a QV of 2^20 is 1,000 times 2^10, not twice. It says little about machines too large to test this way.",
    "origin": {
      "created_by": "IBM researchers (Cross, Bishop, Sheldon and colleagues), 2019",
      "vendor_involvement": "Created by IBM, a hardware maker; now used across the industry."
    },
    "references": [
      "https://arxiv.org/abs/1811.12926"
    ],
    "page_url": "https://qubitbenchmark.com/benchmarks/quantum-volume/"
  },
  {
    "id": "quops",
    "name": "QUOPS",
    "short_name": "QUOPS",
    "kind": "benchmark",
    "level": "system",
    "definition": "How large a random circuit a machine can run successfully (the QUOPS score) and how many such operations it completes per second (the QUOPS rate).",
    "unit": "QUOPS score",
    "direction": "higher_is_better",
    "method_versions": [
      {
        "version": "v1 (Proctor et al. 2026, preprint)",
        "reference_url": "https://arxiv.org/abs/2609.12146"
      }
    ],
    "applies_to": [
      "gate-based machines"
    ],
    "comparability": "Scores are comparable across architectures run to the same protocol. The score and the rate measure different things and should not be combined into one number.",
    "origin": {
      "created_by": "Sandia National Laboratories with Quantinuum and NVIDIA (Proctor et al.), 2026",
      "vendor_involvement": "Quantinuum, a hardware maker, helped design QUOPS."
    },
    "references": [
      "https://arxiv.org/abs/2609.12146"
    ],
    "page_url": "https://qubitbenchmark.com/benchmarks/quops/"
  },
  {
    "id": "clops",
    "name": "Circuit layer operations per second",
    "short_name": "CLOPS",
    "kind": "benchmark",
    "level": "system",
    "definition": "How many layers of a standard test circuit a machine runs per second, including the time spent by the classical computers around it.",
    "unit": "layers per second",
    "direction": "higher_is_better",
    "method_versions": [
      {
        "version": "Wack et al. 2021",
        "reference_url": "https://arxiv.org/abs/2110.14108"
      }
    ],
    "applies_to": [
      "gate-based machines"
    ],
    "comparability": "Depends on the circuit size used and on the cloud software stack, so compare only results that state the same version and set-up.",
    "origin": {
      "created_by": "IBM researchers (Wack, Paik, Javadi-Abhari and colleagues), 2021",
      "vendor_involvement": "Created by IBM, a hardware maker."
    },
    "references": [
      "https://arxiv.org/abs/2110.14108"
    ],
    "page_url": "https://qubitbenchmark.com/benchmarks/clops/"
  },
  {
    "id": "algorithmic-qubits",
    "name": "Algorithmic qubits",
    "short_name": "#AQ",
    "kind": "benchmark",
    "level": "application",
    "definition": "The number of qubits a machine can use successfully across a set of standard algorithms, based on the QED-C application benchmarks.",
    "unit": "algorithmic qubits",
    "direction": "higher_is_better",
    "method_versions": [
      {
        "version": "Based on QED-C benchmarks (Lubinski et al.)",
        "reference_url": "https://arxiv.org/abs/2110.03137"
      }
    ],
    "applies_to": [
      "gate-based machines"
    ],
    "comparability": "Defined and mostly reported by IonQ. The test suite has changed over time, so compare only results that name the same version.",
    "origin": {
      "created_by": "IonQ, building on the QED-C application-oriented benchmarks",
      "vendor_involvement": "Defined by IonQ, a hardware maker."
    },
    "references": [
      "https://arxiv.org/abs/2110.03137"
    ],
    "page_url": "https://qubitbenchmark.com/benchmarks/algorithmic-qubits/"
  },
  {
    "id": "q-score",
    "name": "Q-score",
    "short_name": "Q-score",
    "kind": "benchmark",
    "level": "application",
    "definition": "The largest Max-Cut optimisation problem, in number of variables, that a machine solves noticeably better than random guessing.",
    "unit": "Q-score",
    "direction": "higher_is_better",
    "method_versions": [
      {
        "version": "Martiel et al. 2021",
        "reference_url": "https://arxiv.org/abs/2102.12973"
      }
    ],
    "applies_to": [
      "gate-based machines",
      "annealers"
    ],
    "comparability": "Comparable when run to the published protocol. It tests one kind of optimisation problem, not general ability.",
    "origin": {
      "created_by": "Atos researchers (Martiel, Ayral, Allouche), 2021",
      "vendor_involvement": null
    },
    "references": [
      "https://arxiv.org/abs/2102.12973"
    ],
    "page_url": "https://qubitbenchmark.com/benchmarks/q-score/"
  },
  {
    "id": "xeb",
    "name": "Cross-entropy benchmarking",
    "short_name": "XEB",
    "kind": "benchmark",
    "level": "system",
    "definition": "How closely a machine's output from random circuits matches the ideal, expressed as a fidelity between 0 and 1.",
    "unit": "fidelity",
    "direction": "higher_is_better",
    "method_versions": [
      {
        "version": "Boixo et al. 2018",
        "reference_url": "https://arxiv.org/abs/1608.00263"
      }
    ],
    "applies_to": [
      "gate-based machines"
    ],
    "comparability": "Only comparable for the same number of qubits and circuit depth. Mostly used in quantum-advantage experiments.",
    "origin": {
      "created_by": "Google researchers (Boixo, Isakov, Smelyanskiy and colleagues), 2018",
      "vendor_involvement": "Created by Google, a hardware maker."
    },
    "references": [
      "https://arxiv.org/abs/1608.00263"
    ],
    "page_url": "https://qubitbenchmark.com/benchmarks/xeb/"
  },
  {
    "id": "randomized-benchmarking",
    "name": "Randomized benchmarking",
    "short_name": "RB",
    "kind": "benchmark",
    "level": "component",
    "definition": "The average error per gate, found by running random sequences of gates of growing length and measuring how fast the results decay.",
    "unit": "error per gate",
    "direction": "lower_is_better",
    "method_versions": [
      {
        "version": "Magesan, Gambetta and Emerson 2011",
        "reference_url": "https://arxiv.org/abs/1009.3639"
      }
    ],
    "applies_to": [
      "gate-based machines"
    ],
    "comparability": "Variants (standard, interleaved, simultaneous, direct) give different numbers. Compare only the same variant on the same kind of gate.",
    "origin": {
      "created_by": "Academic research (Knill and colleagues; Magesan, Gambetta and Emerson)",
      "vendor_involvement": null
    },
    "references": [
      "https://arxiv.org/abs/1009.3639"
    ],
    "page_url": "https://qubitbenchmark.com/benchmarks/randomized-benchmarking/"
  },
  {
    "id": "two-qubit-gate-fidelity",
    "name": "Two-qubit gate fidelity",
    "short_name": "2Q fidelity",
    "kind": "measurement",
    "level": "component",
    "definition": "How often an operation on two qubits gives the right result. 99.9% means about one error in a thousand operations.",
    "unit": "%",
    "direction": "higher_is_better",
    "method_versions": [],
    "applies_to": [
      "gate-based machines"
    ],
    "comparability": "Best, median and average values differ a lot; so do single pairs and whole-chip figures. Compare only like with like.",
    "origin": {
      "created_by": "Standard measure in quantum computing",
      "vendor_involvement": null
    },
    "references": [],
    "page_url": "https://qubitbenchmark.com/benchmarks/two-qubit-gate-fidelity/"
  },
  {
    "id": "layer-fidelity",
    "name": "Layer fidelity (error per layered gate)",
    "short_name": "EPLG",
    "kind": "measurement",
    "level": "system",
    "definition": "The error per two-qubit gate when many gates run at once across a chain of qubits, which captures interference between neighbouring qubits.",
    "unit": "error per layered gate",
    "direction": "lower_is_better",
    "method_versions": [
      {
        "version": "McKay et al. 2023",
        "reference_url": "https://arxiv.org/abs/2311.05933"
      }
    ],
    "applies_to": [
      "gate-based machines"
    ],
    "comparability": "Depends on the length of the qubit chain tested (often 100 qubits). Compare only the same chain length.",
    "origin": {
      "created_by": "IBM researchers (McKay, Hincks, Pritchett and colleagues), 2023",
      "vendor_involvement": "Created by IBM, a hardware maker."
    },
    "references": [
      "https://arxiv.org/abs/2311.05933"
    ],
    "page_url": "https://qubitbenchmark.com/benchmarks/layer-fidelity/"
  },
  {
    "id": "logical-error-rate",
    "name": "Logical error rate",
    "short_name": "Logical error",
    "kind": "measurement",
    "level": "system",
    "definition": "How often an error-corrected (logical) qubit still fails, usually per round of error correction.",
    "unit": "error per cycle",
    "direction": "lower_is_better",
    "method_versions": [],
    "applies_to": [
      "machines running error correction"
    ],
    "comparability": "Depends on the code, its size (distance), what counts as one cycle, and whether failed runs are discarded. Compare only identical set-ups.",
    "origin": {
      "created_by": "Standard measure in quantum error correction",
      "vendor_involvement": null
    },
    "references": [],
    "page_url": "https://qubitbenchmark.com/benchmarks/logical-error-rate/"
  },
  {
    "id": "physical-qubit-count",
    "name": "Physical qubits",
    "short_name": "Qubits",
    "kind": "measurement",
    "level": "system",
    "definition": "How many working physical qubits the machine has.",
    "unit": "qubits",
    "direction": "higher_is_better",
    "method_versions": [],
    "applies_to": [
      "all machines"
    ],
    "comparability": "A count says nothing about quality. Gate-based machines and annealers are counted separately, and so are machines and single experiments.",
    "origin": {
      "created_by": "Standard measure",
      "vendor_involvement": null
    },
    "references": [],
    "page_url": "https://qubitbenchmark.com/benchmarks/physical-qubit-count/"
  },
  {
    "id": "logical-qubit-count",
    "name": "Logical qubits",
    "short_name": "Logical qubits",
    "kind": "measurement",
    "level": "system",
    "definition": "How many error-corrected (logical) qubits the machine has run at once, as demonstrated in a published experiment.",
    "unit": "logical qubits",
    "direction": "higher_is_better",
    "method_versions": [],
    "applies_to": [
      "machines running error correction"
    ],
    "comparability": "Only meaningful with the logical error rate and the code used. A logical qubit that fails more often than a physical one is not progress.",
    "origin": {
      "created_by": "Standard measure in quantum error correction",
      "vendor_involvement": null
    },
    "references": [],
    "page_url": "https://qubitbenchmark.com/benchmarks/logical-qubit-count/"
  }
]