Q-CTRL reports cleaner magic states on IBM hardware anyone can rent
Q-CTRL published results on November 18, 2026 from a magic state cultivation experiment run on IBM's publicly accessible Quantum Nighthawk processor, with no privileged access and using the company's performance-management software. Authors Haoran Liao and Yuval Baum describe what the company says is the first experimental demonstration of cultivation with a lattice-surgery escape. Cultivation was introduced in 2024 and the lattice-surgery variant was proposed in 2025.
The cultivated T state had a logical error probability of 2 × 10^-5, a fidelity of 0.99998, while keeping 25% of shots. Google's Willow experiment, the only earlier hardware demonstration, reported 1.9 × 10^-4 with 8.7% of shots retained. Q-CTRL calls that roughly ten times lower error at about three times the acceptance. Right after lattice surgery, fidelity in the surface code passed 0.99 at up to 16% retention, against a prior record near 0.91 in a smaller code. The route lands in a standard surface code that works with minimum-weight perfect matching decoders, which Q-CTRL says process data over 1,000 times faster than the decoder grafted codes need.
Why it matters: magic states are a main bottleneck for fault-tolerant quantum computing, and early estimates had factories eating over 90% of a machine's physical qubits. Cheaper, cleaner states mean fewer qubits spent on overhead and a shorter path to quantum machines useful for materials and science, reachable through public cloud access rather than a few private labs.
Caveats are real. This comes from the company's own blog post and manuscript (arXiv:2611.22035), with no independent verification. The fidelities depend on postselection, discarding most shots. The demonstration is at distance 5 only, and the Google comparison is approximate because the circuits are similar, not identical. Still to come are larger code distances on bigger hardware, and using the states in logical T gates with real-time decoding.
Source: Google News
Many Minded