French quantum computing company Alice & Bob has achieved a significant milestone in quantum error correction, completing GPU-accelerated decoding of its proprietary Elevator Codes in under two hours.
Through integration with NVIDIA CUDA-Q, Alice & Bob achieved a runtime of 1 hour and 57 minutes for the decoding of simulated syndrome data from its Elevator Codes, a concatenation-based error correction architecture developed specifically for biased noise cat qubits.
What Are Elevator Codes?
Elevator Codes are a new error correction architecture that modifies the active protection of the repetition code to tackle bit-flips as well as phase-flips. The company’s cat qubits are partially protected from bit-flip errors by design, and the Elevator Codes extend this protection to achieve significantly lower logical error rates.
Research published earlier in 2026 demonstrated that Elevator Codes could achieve up to 10,000 times lower logical error rates with only a modest increase in the number of required qubits. The codes were described in a pre-print by Diego Ruiz, a theoretical physicist at Alice & Bob, and Peter Shanahan, a quantum informatics researcher.
GPU Acceleration as a Key Enabler
The sub-two-hour decoding milestone demonstrates that classical computing acceleration, when paired with quantum error correction frameworks, can handle the computational demands of realistic error correction workloads. GPU-accelerated simulation is increasingly viewed as essential for validating quantum error correction protocols before they are deployed on physical hardware.
Alice & Bob’s cat qubit approach represents a distinct path in the quantum computing landscape, offering inherent noise protection at the hardware level that reduces the overhead typically required by error correction codes. The integration with NVIDIA CUDA-Q provides a scalable pathway for simulating increasingly complex error correction scenarios.
