The Paris based company calls Helium the first multi cat qubit system made available to outside organizations. External teams can now run hardware efficient fault tolerance experiments on real cat hardware, inside a tightly published ceiling.
Alice & Bob has done something quietly significant in the quantum error correction (QEC) research landscape: it has opened Helium, its 18-cat-qubit processor, to third-party researchers for QEC experiments. The company describes Helium as the first multi cat-qubit system made available to outside organizations, a shift from in-house demonstrations to reproducible, external experimentation on a noise-biased qubit modality.
The launch was summarized in The Qubit Report's coverage of the announcement, which consolidates the platform specs, funding context, and roadmap. The underlying technical story, however, traces to Alice & Bob's own posts, including its January 2026 'Elevator Codes' writeup on the bit-flip performance and the April 2026 'Scaling the Cats' post on the Paris product-development lab that makes the access model physically possible.
What external researchers actually get is specific. Helium centers on an 18 noise-biased cat-qubit chip with degree-2 connectivity, paired with a 10 mK processor stage inside a custom cryostat. The full rack is 2.4 m × 3.3 m × 9 m and occupies roughly 21 m², drawing about 40 kW. Ambient requirements are 18–25 °C, 45–65% relative humidity, and acoustic levels under 60 dB. That is not a desktop experiment; it is a system sized for a research-grade lab, the kind of hardware external teams book time on rather than plug in.
The headline numbers define the experimental ceiling. Alice & Bob reports bit-flip times greater than one hour and a 94.2% Z-gate fidelity executed in 26.5 nanoseconds, figures that translate the Elevator Codes result into a packaged spec. The bit-flip performance is the substantive point of cat qubits: by biasing the noise channel, the architecture suppresses one error class by design, which is the basis of Alice & Bob's claim that the approach can cut fault-tolerant quantum computing (FTQC) hardware overhead by up to 200× relative to conventional superconducting platforms. The catch, named in the same announcement, is that 94.2% sits well below the >99% two-qubit-class fidelity that most fault-tolerance schemes ultimately require, and degree-2 connectivity constrains which code topologies can actually be compiled onto the chip.
For third-party work, the current scope is bounded. Helium targets one logical qubit via repetition codes, with a 10^-2 logical error rate as the working target. That target is high relative to surface-code thresholds, and positions the platform as a research testbed for hardware-efficient QEC strategies rather than a system on which to run deep, near-term applications. What researchers can do is stress-test repetition-code decoding against real cat-qubit noise, benchmark suppression strategies on a multi-qubit array, and produce comparable, reproducible results on a noise-biased modality. Until now, that class of experiment has largely been the company's own, run on its own hardware, reported on its own terms.
The access regime is the part worth naming precisely. Third-party access is mediated by Alice & Bob, not opened as an unrestricted public cloud service. The April 2026 'Scaling the Cats' post frames the new Paris product-development lab as the engineering reason this model is now possible: a roughly 4,000 m² facility sized for up to 20 Bluefors dilution refrigerators and hybrid control stacks from Quantum Machines, with environmental isolation designed so the company can ship customer-installable racks rather than one-off demonstrations. The thesis is reproducibility as a product feature, not a research afterthought.
The funding context reinforces that this is an infrastructure bet, not a single-product story. Alice & Bob has raised roughly €130M (about $142M) to date, and the Platform page roadmap lays out a four-step path: Helium (now), then Lithium, Beryllium, and finally Graphene, a system the company targets at 100 logical qubits by 2030. Each step is meant to add logical-qubit capacity while preserving the same noise-biasing premise. If Helium is the first externally testable node, Graphene is the destination the Paris lab is being built to enable.
What to watch next is whether the third-party program produces independent results that confirm or complicate the company's numbers on real cat-qubit hardware, and whether degree-2 connectivity and the 94.2% fidelity floor become binding constraints as the roadmap moves from one logical qubit toward the 100-logical-qubit target. The platform is now open, the ceiling is published, and the field has a multi-qubit cat array to test against.