Quantum error correction in 2026: below threshold, still far from useful
· TexaHub · Quantum, Technology Trends
Several platforms now show encoded qubits outperforming physical ones, but the logical qubit counts and error rates demonstrated remain far from what useful fault-tolerant algorithms require.
What the terms mean
Quantum error correction combines many error-prone physical qubits into one logical qubit, and in principle the logical error rate falls exponentially as more physical qubits are added [2]. That suppression only happens if the physical error rate sits below a critical threshold; above it, adding qubits adds more errors than the code can remove [2]. A related benchmark is break-even: the point at which an encoded qubit outlasts or outperforms the best unencoded qubit on the same device [2].
Being below threshold is therefore a precondition for scaling rather than a measure of usefulness. A logical qubit that improves with size can still be far too error-prone, too slow or too expensive in hardware for practical algorithms.
The benchmark results
Google's December 2024 Willow result remains the reference point for superconducting qubits: on the 105-qubit chip, the company reported that each step from a 3x3 to a 5x5 to a 7x7 grid of encoded qubits cut the error rate in half [1]. The accompanying paper reports a suppression factor of 2.14 per increase in code distance, a 101-qubit distance-7 surface code with 0.143% error per correction cycle, and a logical lifetime 2.4 times that of the best physical qubit [2]. The same paper notes that performance was limited by rare correlated error events occurring roughly once an hour [2].
Neutral atoms reached a comparable milestone. A Harvard, MIT and QuEra team used arrays of up to 448 atoms to show 2.14-times below-threshold performance in a surface-code circuit, along with logical entanglement, universal logic via teleportation and deep circuits with dozens of logical qubits [3].
Trapped ions have pushed on efficiency. Quantinuum says its 98-qubit Helios system, launched in November 2025 with 99.921% two-qubit gate fidelity, produced 48 error-corrected logical qubits at a 2:1 encoding rate [4]. In September 2026 the company reported that its Helix architecture, built by concatenating two smaller codes, achieved a logical memory error of 4.6 x 10^-5 per qubit per round without post-selection, and a logical Clifford gate error of 2.8 x 10^-4, which it describes as a 4.28-times improvement on Helios' physical gate error [5].
Cheaper codes and different qubits
Much of the current work targets overhead, the number of physical qubits each logical qubit costs. IBM says its bivariate bicycle "gross" code corrects errors as well as the surface code while requiring 10 times fewer qubits [6]. In November 2025 IBM announced its experimental Loon processor and said it had decoded qLDPC-code errors in real time in under 480 nanoseconds, a year ahead of schedule [7].
Amazon's Ocelot chip uses cat qubits whose bit-flip times approach one second, leaving mainly phase flips to correct [8]. Its logical error rate was 1.72% per cycle at distance 3 and 1.65% at distance 5, and Amazon projects that scaling the design could cut error-correction overhead by up to 90% compared with surface codes [8]. Microsoft is pursuing topological qubits; in June 2026 it reported a mean qubit lifetime of 20 seconds for its Majorana 2 chip, a physical-qubit result rather than a logical-qubit demonstration [9].
How far from useful machines
The gap is still large. IBM's own target, attributed to its roadmap, is Starling by 2029: a machine running circuits of 100 million gates on 200 logical qubits, preceded by the Kookaburra module scheduled for 2026 [6]. Microsoft says it now expects a scalable quantum computer by 2029 [9], and Quantinuum describes Helix as the foundation for its next system, Apollo [5].
For a concrete workload, Craig Gidney estimated in 2025 that factoring a 2048-bit RSA integer would take less than a week on a machine with less than a million noisy qubits, assuming a 0.1% gate error rate [10]. The Willow, Helios and Harvard experiments used between 98 and 448 physical qubits [1] [3] [4], so that estimate implies roughly three orders of magnitude of scaling, even with error correction now behaving as theory predicts.
Sources
[1] Meet Willow, our state-of-the-art quantum chip — Google, 9 Dec 2024
[2] Quantum error correction below the surface code threshold — arXiv (Google Quantum AI and collaborators), 24 Aug 2024
[3] Architectural mechanisms of a universal fault-tolerant quantum computer — arXiv (Bluvstein et al., Harvard/MIT/QuEra), 25 Jun 2025
[4] Introducing Helios: The Most Accurate Quantum Computer in the World — Quantinuum, 5 Nov 2025
[5] Helix: A New Architecture for Enterprise-Scale Fault-Tolerant Quantum Computing — Quantinuum, 8 Sep 2026
[6] IBM lays out clear path to fault-tolerant quantum computing — IBM Quantum, 10 Jun 2025
[7] IBM Delivers New Quantum Processors, Software, and Algorithm Breakthroughs on Path to Advantage and Fault Tolerance — IBM Newsroom, 12 Nov 2025
[8] Amazon announces Ocelot quantum chip — Amazon Science, 27 Feb 2025
[9] Majorana 2, made more reliable with Microsoft Discovery agentic AI — Microsoft Source, 2 Jun 2026
[10] How to factor 2048 bit RSA integers with less than a million noisy qubits — arXiv (Craig Gidney), 21 May 2025