How we rank

We rank races, not a winner

There's no single "best quantum computer". A machine can lead on accuracy and trail on size, and rolling everything into one score would hide that. So we run 5 separate races, each built on one measure:

• Most accurate: two-qubit gate fidelity.
• Error-corrected qubits: logical qubits demonstrated.
• Most qubits: physical qubits, gate-based machines.
• Fastest: circuit layer operations per second.
• Use it today: largest machine on a public cloud.

Where the numbers come from

Every number comes from QubitBenchmark, our sister site, which records each published result with its source, its dates, who measured it, and how it was checked. QubitRank keeps no numbers of its own.

QubitBenchmark publishes one checked data release each month. Every ranking page names the release it uses (this month: 2026-11.1), and every rank links to the exact record behind it. Old releases never change, so any past ranking can be rebuilt and checked.

Which results count

QubitBenchmark grades every result:

• A Independent and peer-reviewed. Checked by people outside the company. A peer-reviewed paper by the machine's own maker gets a B at most.
• B Documented. A preprint or technical report that explains how the result was measured.
• C: Announcement only. A press release or claim without the method. Grade C never decides a rank. We may report it as news, clearly labelled.

A result only counts if QubitBenchmark marks it as eligible: grade A or B, still current (not corrected, replaced, withdrawn or disputed), and with a working source. Every category page also lists the results that didn't count, and why.

Comparing like with like

Two results are only ranked against each other if QubitBenchmark has judged them comparable: the same measure, taken the same way. Results measured differently stay out of the same race, even if the numbers look similar. Where different kinds of machine aren't a fair comparison, the category page says so.

One result per machine

• Each machine is ranked on its latest eligible result in that race.
• Equal results share a place.
• Retired machines leave the current rankings, but their history stays on QubitBenchmark.
• "Most qubits" covers gate-based machines only. Quantum annealers work differently and would make the comparison misleading.
• "Use it today" counts only machines QubitBenchmark records as available on a public cloud, as of the date it last checked.

What moves a rank, and what never does

• Ranks change once a month, when the new data release is published. ▲▼ show the change since last month; "New" means the machine wasn't ranked last month.
• A press release, a headline or a conference talk never moves a rank on its own.
• No company can pay to be listed, ranked or placed higher. We don't accept payment from anyone to influence a result or a rank.
• Rankings describe technology, not investments, and aren't financial advice.

Other organisations' rankings

When another organisation publishes its own ranking or results, we may report it in our news, credited and linked, in our own words. It's shown separately and never mixed into our ranks.

Mistakes and corrections

If a result is wrong, QubitBenchmark corrects it in a new release and keeps a record of the change. Our ranking changes with it, and we never quietly rewrite a past ranking. If you spot an error, tell us and we'll check it against the source.