Gate fidelity measures how accurately a real quantum gate implements its ideal mathematical operation. A fidelity of 99.9% (often written as "three nines") means the gate introduces an error approximately 0.1% of the time. This seemingly small error rate compounds rapidly across a circuit — a 100-gate circuit with 99.9% fidelity per gate has an overall success probability of roughly 90%, while a 1,000-gate circuit drops to about 37%.
Gate fidelity is typically measured through randomized benchmarking (RB), which applies random sequences of Clifford gates followed by an inversion gate and measures how the error accumulates with sequence length. This technique isolates gate errors from state preparation and measurement (SPAM) errors. Cross-entropy benchmarking (XEB), used by Google, provides a complementary metric by comparing the output distribution of random circuits against classical simulation.
Current state-of-the-art fidelities across platforms: superconducting transmon single-qubit gates achieve 99.95-99.99%, two-qubit gates 99.5-99.9% (Google Willow, IBM Heron); trapped-ion single-qubit gates exceed 99.999%, two-qubit gates reach 99.7-99.9% (Quantinuum H2); neutral atom two-qubit Rydberg gates achieve 99.5% and improving (QuEra). The surface code error threshold of approximately 1% means processors need two-qubit gate fidelities above 99% for error correction to function, a bar now met by multiple platforms.