Quantinuum announced a new quantum error-correction record on October 19: its commercial Helios quantum computer achieved 99.9995% logical fidelity using a *color code* — five errors per million logical operations.
To put the number in everyday terms: a quantum computer performing one million operations per second at this error rate would average one error every 0.2 seconds, while practical quantum algorithms may require trillions of operations. Three decades of error-correction research have been a slow battle to push error rates down, and 99.9995% is the latest step.
What makes the announcement notable is where the number came from: the same machine set a record two months earlier using a different code. Same hardware, different code, error rate down another tenfold.
🏆 The record
Quantinuum says the record was achieved on Helios, a commercially available machine that customers can rent, using the color code. The company offers two comparisons:
- 10x lower error rate than its own September record, set with the Helix code
- 100x lower than the best publicly reported result from any other technical route, achieved by Google Quantum AI on a superconducting system
- September record (Helix code): 10 physical qubits per logical qubit
- October record (color code): up to 37 physical qubits per logical qubit
- All figures come from the company press release; internal details of the commercial system have not been peer-reviewed.
- Blume-Kohout's "one million physical qubits" is an estimate about a scaled-up system, not a description of this experiment.
- The color code achieves lower logical errors but costs several times more physical qubits per logical qubit than the Helix code; Quantinuum says it is still weighing which code suits which task.
The long-running ion-trap vs. superconducting rivalry thus has a partial answer: a 100x gap between the best public results of the two leading players.
🔧 Two months, two codes
The color code costs more physical qubits in exchange for a lower logical error rate.
The experiment also produced the curve error-correction theorists most want to see: each step up in code distance reduced residual logical errors by a record factor of ~12x. Exponential error suppression with increasing code distance means the route follows the correct scaling law.
> Note: Code distance can be thought of as the "thickness of the protection layer" — like making more copies of a document to compare against each other. In quantum error correction, each increment in code distance raises physical-qubit overhead roughly quadratically while reducing the logical error rate by a multiplicative factor.
🧭 What comes after 100x
CEO Rajeeb Hazra said the result demonstrates that Quantinuum has the error-correction architecture and ion-trap technology needed for universal fault tolerance, pointing toward Apollo, a general-purpose fault-tolerant computer planned for 2029.
A more cautious third-party view came from Robin Blume-Kohout, co-director of the Quantum Performance Laboratory at Sandia National Laboratories (not involved in the research): the paper shows real, serious error-correction gains — a logical qubit nearly 100x better than the physical qubits composing it — and scaling this system to the reliability needed for useful fault-tolerant computing could stay under one million physical qubits, improving the outlook for practical quantum computing.
| Comparison | September, Helix code | October, color code | |---|---|---| | Logical fidelity | Record at the time | 99.9995% | | Physical qubits per logical qubit | 10 | Up to 37 | | vs. previous month | Baseline | Error rate down another 10x | | Per code-distance step | Not disclosed | Logical errors drop ~12x |