Quantinuum achieves 99.9995% logical fidelity: the record that redefines quantum computing

On October 19, Quantinuum announced the highest quantum error correction fidelity ever recorded: a logical fidelity of 99.9995% on the Helios computer, equivalent to just five errors per million operations. The achievement, published simultaneously as a technical study and press release, represents not merely a number but a conceptual milestone — trapped-ion technology distances itself from superconducting competition by a factor of 100 in logical fidelity, and from Quantinuum's own previous record by a factor of 10.

What is logical fidelity and why it matters

In quantum computing, "fidelity" measures how faithfully a system preserves quantum information during an operation. Physical qubits — the individual units of quantum information — are extremely fragile. Any interaction with the environment, whether temperature, electromagnetic radiation, or mechanical vibration, can corrupt their state. Quantum error correction (QEC) attempts to solve this by encoding one logical qubit across multiple physical qubits, so that individual errors can be detected and corrected without destroying the quantum information.

The 99.9995% logical fidelity means that in each "syndrome extraction cycle" — a step of the error correction protocol — the probability of an uncorrectable error is only 5 in a million. To put that in perspective: the previous best result from superconducting systems, reported by Google Quantum AI, was 100 times worse. This is not a marginal difference; it is a scale difference that separates what is experimental from what is potentially usable.

The architecture behind the record

The Helios is a 137-barium-137 (¹³⁷Ba⁺) trapped-ion machine organized in a QCCD (quantum charge-coupled device) architecture. The record specifically used a triangular color code — a class of topological error-correcting code — at three code distances: d=3 (13 qubits per logical block), d=5 (37 qubits), and d=7 (73 qubits). The 99.9995% record corresponds to distance 7.

The most significant result, however, is not the isolated number, but the exponential suppression rate: each increase in code distance reduced logical errors by an average factor of 12. In practical terms, this means that each additional layer of protection makes correction significantly more effective — exactly the theoretically expected behavior for quantum error correction, but still difficult to demonstrate experimentally.

What makes this particularly impressive is that Quantinuum achieved two records in two months using the same hardware with different codes. In September, with the Helix code, it demonstrated record-setting logical memory and computation; in October, with the color code, it achieved record logical fidelity. CEO Dr. Rajeeb Hazra describes this as validation of an "error-correction architecture" that allows switching between codes without changing hardware.

Trapped-ion vs. superconducting: a dispute that is redefining itself

The result raises a question the industry has debated for years: which physical qubit platform is best for universal quantum computing? Quantinuum argues that trapped-ion offers fundamental advantages. Trapped-ion qubits have naturally long coherence times — the company has already demonstrated quantum memories lasting over 20 seconds — and all-to-all qubit connectivity, which eliminates the need for swap operations that introduce additional errors.

Google Quantum AI, by contrast, uses superconducting qubits, which are easier to manufacture at large scale but have much shorter coherence times and limited connectivity (each qubit only interacts directly with neighboring qubits). Google's previous record, now surpassed by a factor of 100, was already considered impressive — but the comparison between platforms with such different architectures raises the question of whether the trapped-ion advantage will hold as both technologies scale.

Quantinuum projects that, at the measured suppression factor of Λ=12, target logical error rates of 10⁻¹² across 100 logical qubits can be achieved with approximately 30,000 physical qubits — a number the company considers "feasible" within its production line. Google's estimate for its superconducting platform, with similar parameters, would require a significantly higher physical qubit count.

The path to Apollo and error correction below threshold

The Apollo project, the universal fault-tolerant computer slated for 2029, is the stated goal of this series of records. But the fidelity record is only part of the story. A parallel study, presented on arXiv (2602.22211), demonstrates computations with 48 to 94 logical qubits using high-rate codes that exceed the "break-even point" — the point at which error correction actually pays off, i.e., when the logical qubit becomes more reliable than the best individual physical qubit.

This is the true conceptual milestone: for years, quantum error correction was theoretically proven but experimentally, each additional layer of protection also introduced more errors into the correction process. Break-even was the point at which the balance turned positive. What Quantinuum demonstrated in September and October is that this point has already been surpassed in multiple configurations — memory, computation, and gate — on the same hardware.

What remains to be done

Despite the advances, the distance between what has been achieved and a fully functional universal quantum machine remains significant. A 100-logical-qubit system with 30,000 physical ones still requires real-time syndrome error control with latency below 0.5 milliseconds — something the Quantinuum study showed is compatible with neural network decoders on NVIDIA H100 hardware, but which operates at the edge of what is currently practical. Furthermore, distance-7 codes use 73 physical qubits per logical block; a 100-logical-qubit system would require multiple blocks, each with its own correction overhead.

Quantum error correction well below the color code threshold — the title of Quantinuum's technical study — is without doubt a notable scientific achievement. But universal quantum computing, capable of executing algorithms like Shor's or complex chemical simulation, still requires advances in scalability, gate control, and real-time decoding that, while now arguably viable, are not yet solved.

The question is whether the 99.9995% logical fidelity will be remembered as a destination — or merely as a stepping stone — toward a quantum machine that truly changes the game.

Sources: Quantinuum, Quantum Computing Report, arXiv

✓ Independent sources cross-checked and verified before publishing