NVIDIA Launches Ising, the First Open-Source AI Models for Quantum Computing
NVIDIA releases Ising, an open-source family of AI models that automates qubit calibration and runs real-time quantum error decoding 2.5x faster than the pyMatching standard.
NVIDIA has unveiled Ising, the world’s first family of open-source AI models built specifically to accelerate the path to useful, fault-tolerant quantum computers. Released on April 14, 2026, the models tackle two of the most stubborn bottlenecks in quantum hardware: continuous calibration of finicky qubits and real-time decoding for quantum error correction.
The release includes two distinct components. Ising Calibration is a vision-language model that interprets multi-modal qubit measurements and automates calibration workflows that previously took quantum engineers days of manual tuning per processor. NVIDIA says agents built on the model can now keep machines tuned in hours instead of days. Ising Decoding ships in two 3D convolutional neural network variants — one optimized for raw speed, the other for accuracy — and runs in the real-time loop that catches and corrects quantum errors as they occur.
The performance numbers are eye-catching. NVIDIA reports Ising Decoding is up to 2.5x faster and 3x more accurate than pyMatching, the open-source decoder that has been the de facto industry standard. That speed margin matters because quantum error correction has to keep up with qubit decoherence in real time; a slower decoder means accumulated errors that wreck a computation before it finishes.
A who’s-who of quantum research labs has already adopted the models. Early users include Academia Sinica, Fermi National Accelerator Laboratory, Harvard’s Paulson School of Engineering, Infleqtion, IQM Quantum Computers, Lawrence Berkeley National Laboratory’s Advanced Quantum Testbed, and the U.K. National Physical Laboratory. The breadth of academic and industrial uptake suggests Ising could become a common substrate for the field rather than a vendor-locked stack.
Ising integrates with the NVIDIA CUDA-Q hybrid quantum-classical platform and the new NVQLink QPU-to-GPU interconnect, both designed for the tight latency budgets that error correction demands. The models are available on GitHub, Hugging Face, and build.nvidia.com, with NIM microservices for fine-tuning. By open-sourcing the stack rather than locking it inside a proprietary cloud, NVIDIA is betting that quantum’s growing pains will produce more useful tools faster — and that owning the GPU layer underneath every quantum control system is the more durable position.
