Caltech and Oratomic's "mitten codes" cut the qubit cost of quantum error correction by roughly 100x

Quantum error correction is the translator layer that lets a fragile quantum processor behave like a reliable computer, and the cost of that translation is the field's main scaling tax. Caltech and Oratomic's "mitten codes" shrink that tax by roughly two orders of magnitude on a specific benchmark, in simulation rather than on hardware.

Mitten codes are quantum low-density parity-check (qLDPC) codes built over non-abelian groups, evading the distance-6 ceiling that limits comparable abelian codes. They hold a constant 20% encoding rate at a check weight of 9, and reach code distances of 18 to 24 with only a few hundred physical data qubits. The architecture also supports parallel lattice surgery and parallel magic-state injection, both required operations on the road to a working fault-tolerant machine.

The headline number comes from a memory simulation under circuit-level depolarizing noise. At a 0.4% physical error rate, a mitten code using about 975 physical data qubits to encode 195 logical qubits hits a logical error rate of 10⁻⁸ per syndrome round. A benchmark stack of 195 rotated surface codes reaches the same target with over 100,000 physical qubits, nearly two orders of magnitude more on both qubit count and error rate (Quantum Computing Report).

The simulation runs below current superconducting hardware error rates, so the result benchmarks an architecture rather than a deployable machine. No independent hardware replication has been reported. The arXiv preprint is the primary signal; the next question is whether the same overhead savings hold up under more realistic noise models and on a chip.