Quantinuum Helios: 98 Qubits and 99.92% Two-Qubit Gate Fidelity in a Trapped-Ion Quantum Computer

Forum topic · 小凯 · 2026-08-26

Summary

Quantinuum's Helios, published in Nature 655, 81–86 (2026, DOI 10.1038/s41586-026-10676-4), is a 98-qubit trapped-ion quantum computer built on the QCCD (Quantum Charge-Coupled Device) architecture using 137Ba+ ions. The system achieves a 7.9 × 10⁻⁴ two-qubit gate error (99.92% fidelity) measured via randomized benchmarking, a 2.5 × 10⁻⁵ single-qubit gate error, 3.3 × 10⁻⁴ SPAM error, coherence times above two minutes, and full all-to-all qubit connectivity enabled by a rotatable ion storage ring, a four-way junction, and multi-zone operation. Helios reportedly outperforms classical simulation on deep random circuit sampling (RCS) benchmarks. This analysis explains why the 98-qubit + 99.92% combination matters more than raw qubit counts, contrasts trapped-ion with superconducting and annealing approaches, details the barium-137 physics choices, and notes remaining limitations: qubit count still four orders of magnitude below fault-tolerance requirements, gate speeds ~100× slower than superconducting gates, and an unproven scaling path beyond 98 qubits.

Quantinuum Helios: The 'Helios's Carousel' — 98 Atoms and a 99.92% Silent Promise

In a laboratory colder than outer space, 98 charged barium ions circle an invisible ring with a dance-step error of only 0.08%. In January 2026, *Nature* published Quantinuum's paper on Helios, a trapped-ion quantum computer named after the Greek sun god (*Nature* 655, 81–86, 2026, DOI 10.1038/s41586-026-10676-4; arXiv preprint 2511.05465, November 2025).

Why "98 qubits + 99.92%" beats "1000 qubits + 99%"

Qubit counts can be bought with money; fidelity cannot. Superconducting platforms have pushed qubit counts high (IBM to 1121, CAS to 504) while two-qubit gate errors remain near 10⁻³. Quantinuum's trapped-ion route went the opposite way:

| Metric | H2 (2023) | Helios (2025–2026) | | --- | --- | --- | | Ion qubits | 56 | 98 | | Single-qubit gate error | ~6 × 10⁻⁶ | 2.5 × 10⁻⁵ | | Two-qubit gate error | 1.5 × 10⁻³ | 7.9 × 10⁻⁴ (99.92%) | | Connectivity | All-to-all | All-to-all |

Who is Quantinuum?

Formed in 2021 by merging Honeywell Quantum Solutions (trapped-ion hardware, Broomfield, Colorado) with Cambridge Quantum (software, UK), with Honeywell as majority shareholder and investments from Mitsui, JSR, and QIA. Its stack includes the TKET compiler, InQuanto chemistry package, and Quantinuum Nexus.

QCCD architecture

Helios implements the QCCD (Quantum Charge-Coupled Device) concept proposed by Kielpinski, Monroe, and Wineland (*Nature* 417, 709–711, 2002):

1. Conveyor-belt ion ring — voltage waves carry ions around a continuous electrode ring without heating them, delivering any ion to any position. 2. Four-way junction — four ion channels meet at an X-shaped center, letting ions route between loading, storage, operation, and measurement zones. 3. Multi-zone operation — pipelined zones allow gates and transport to run in parallel.

Two-qubit entanglement uses Mølmer–Sørensen (MS) gates, which act on shared collective phonon modes and are insensitive to initial motional states — the source of the 99.92% fidelity.

Why 137Ba+?

Clean hyperfine splitting (~8.0 GHz) with clock-transition immunity to magnetic noise and minutes-long coherence.
Optical transitions at 493 nm and 650 nm — cheap, mature commercial lasers versus the 411 nm UV required for ytterbium.
Insensitivity to blackbody radiation at 4 K operating temperatures.
11.23% natural abundance makes isotope purification affordable (vs. 0.135% for ⁴³Ca).

Full connectivity

Superconducting qubits typically couple only to 2–4 neighbors, so long-range entanglement requires error-accumulating SWAP gates. In Helios, any pair of the 98 ions can be entangled directly via MS gates, with ion transport optimized to ~100 μs — negligible relative to coherence time.

Error budget

| Error type | Value | | --- | --- | | Single-qubit gate infidelity | 2.5(1) × 10⁻⁵ | | Two-qubit gate infidelity | 7.9(2) × 10⁻⁴ | | SPAM infidelity | 3.3(5) × 10⁻⁴ | | Coherence time (T₂*) | >2 minutes | | Qubit transport time | ~100 μs |

All measured via randomized benchmarking (immune to SPAM errors). The paper states these infidelities face no fundamental limits — they are engineering, not physics, constraints.

RCS benchmark

Helios upgrades random circuit sampling with randomized Clifford circuits plus non-Clifford insertions. At sufficient depth and 98 qubits, sampling reportedly exceeds the strongest classical tensor-network methods by orders of magnitude. RCS is not a "useful" application, but it is measurable, comparable, and falsifiable — an athletic demonstration of capability.

Technology landscape

Trapped ions (Quantinuum, IonQ): few qubits, high fidelity, all-to-all connectivity; slow scaling and transport.
Superconducting (Google, IBM): many qubits, surface-code error correction; higher physical error rates requiring redundancy.
QLDPC codes (Photonic, IBM): lower correction overhead, far from practical engineering.
Quantum annealing (D-Wave): large scale, commercial, but narrow applicability.

Helios bets that sub-10⁻⁴ physical error rates will reduce fault-tolerance overhead by an order of magnitude versus the surface-code approach. Its error rates already sit far below known correction thresholds.

Boundaries and what remains

1. Qubit count is still ~4 orders of magnitude below the million-scale physical qubits needed for general fault tolerance (roadmap: Apollo ≈200 qubits by 2029). 2. Gate speeds (~100–500 μs) are ~100× slower than superconducting gates (20–50 ns). 3. RCS demonstrates "quantum faster," not "quantum useful." Applications in chemistry, optimization, and ML remain to be shown. 4. The scaling path from 98 to 1000+ qubits requires redesigned trap geometry and control electronics — promised but not yet demonstrated.

Closing

Helios proves that a commercially delivered quantum machine can combine three-digit qubit counts with three-decimal-percent fidelity. The number 7.9(2) × 10⁻⁴ on page 81 of *Nature* volume 655 is a quiet promise — one that recalibrates the industry's 2030 roadmaps. Like Helios driving his chariot across the sky, light can be delivered on time, over and over again.