Quantinuum has published the full specifications of its Helios quantum processor in Nature, a 98-qubit trapped-ion system that achieves the highest two-qubit gate fidelity of any commercial quantum computer and demonstrates all-to-all connectivity, meaning any pair of its 98 qubits can perform a two-qubit operation directly.
The paper, with 166 co-authors led by Anthony Ransford at Quantinuum and partners at Sandia National Laboratories, represents a generational leap from the company’s previous H2 processor (32 qubits, 2023) and establishes a new benchmark for what a trapped-ion quantum computer can do.
The architecture
Helios is built on the Quantum Charge-Coupled Device (QCCD) architecture, first proposed in 2002 by researchers at NIST. Unlike superconducting quantum processors (IBM, Google), where qubits sit in fixed positions and distant qubits communicate through multi-step SWAP gate chains, trapped-ion QCCD systems physically transport ions between processing zones.
The processor uses 98 barium ions (¹³⁷Ba⁺) as qubits, with ytterbium ions for sympathetic cooling. Barium was chosen because its electronic transitions fall in the visible spectrum (515 nm gates, 493 nm cooling) rather than the ultraviolet wavelengths required for the ytterbium ions used in earlier Quantinuum systems, visible lasers are cheaper, more stable, and more reliable.
The trap is a 15.3 mm chip with a four-way “X” junction connecting a ring storage zone to two linear logic sections. Ions are shuttled through this junction dynamically, qubits not being operated on wait in the ring (random-access memory) or leg (first-in-first-out) storage zones, while up to 16 qubits are processed per batch in one of eight operation zones. The runtime compiler calculates the optimal sorting order to minimize transport overhead.
Crucially, the entire system draws less than 40 kilowatts of power, comparable to a powerful workstation, and uses only 2.8 electrical signals per qubit, down from 9.9 on the company’s first-generation H1 system.
What all-to-all connectivity means
In superconducting quantum processors, qubits can only interact with their nearest neighbors. Connecting distant qubits requires a chain of SWAP gates that adds circuit depth and accumulates errors. The heavy-hex topology of IBM’s recent chips and Google’s Willow grid both face this constraint.
Helios, by contrast, can bring any two qubits physically together for a direct two-qubit gate. This is critical for quantum error correction: all-to-all connectivity enables high-rate quantum error correction codes, such as low-density parity check (LDPC) codes and concatenated codes, that would require prohibitive overhead on fixed-topology chips. Quantinuum achieved 48 logical qubits from 98 physical qubits, a 2:1 encoding ratio. Google’s Willow, by comparison, demonstrated 1 logical qubit from 105 physical, a 105:1 ratio using the surface code.
Record fidelities
The paper reports two-qubit gate fidelity of 99.921 percent, averaged across all operational zones, meaning the system reports the average performance over the entire chip, not just its best pairs. Single-qubit gates achieve 99.9975 percent fidelity. State preparation and measurement (SPAM) reaches 99.967 percent.
The dominant error source is laser phase noise and spontaneous emission, not fundamental physical limits, meaning further improvements are expected as engineering refines the laser stabilization and motional control.
For random circuit sampling (RCS) on 98 qubits at depth 26, the paper estimates that classical simulation would require over 10⁶ years on an exascale supercomputer. Helios sampling fidelity was measured at approximately 3.5 percent, well above the regime where classical simulation is tractable.
Trapped versus superconducting
Helios’ two-qubit gates take about 70 microseconds, roughly 1,000 times slower than the 50,100 nanosecond gates of superconducting systems. But the slower gate speed is compensated by higher fidelity per gate and the elimination of SWAP overhead, so many algorithms actually finish faster in wall-clock time.
Crystal Noel of Duke University, writing an accompanying News & Views article in Nature, notes that “tackling ongoing engineering challenges will be essential for trapped-ion architectures to reach the next frontier”, the path from 98 qubits to the millions needed for full-scale fault-tolerant quantum computing is not yet demonstrated, and the engineering complexity of shuttling ions through junctions at scale remains significant.
What it can do
Quantinuum has already demonstrated applications including simulations of high-temperature superconductivity and 3D quantum magnetism using 48 logical qubits with error correction beyond the break-even threshold. Enterprise partners, BMW Group (fuel cell catalysts), JPMorganChase (post-quantum cryptography), Amgen (biologics discovery), and SoftBank (battery materials), are using the system through cloud access and on-premise deployments.
The paper notes that none of the demonstrated error rates are “fundamentally limited”, meaning the architecture has headroom for continued improvement as engineering refines the control systems.
Source: Ransford, A., Allman, M.S., Arkinstall, J. et al. “A 98-qubit trapped-ion quantum computer with all-to-all connectivity.” Nature (2026). DOI: 10.1038/s41586-026-10676-4