Publication date: 16 April 2026
NVIDIA has launched Ising, a family of open-source AI models aimed at one of the hardest problems in quantum computing: making quantum processors reliable enough for useful work. Announced on 14 April 2026, the release targets two specific bottlenecks, calibration and quantum error correction, and ships with integrations for NVIDIA’s CUDA-Q platform and NVQLink interconnect.
What the Ising models do
The Ising family is split into two components. Ising Calibration is a vision language model that reads measurement data from quantum processors and reacts in near real time, letting AI agents automate the tuning steps that normally occupy research teams for days. Ising Decoding is a 3D convolutional neural network delivered in two variants, one optimised for raw speed and one for accuracy, and handles real-time decoding for quantum error correction.
Together, the two models are positioned as a control plane for quantum hardware, taking over the repetitive measurement and correction work that sits between a physical qubit and a usable logical qubit.
Technical specs and performance
According to NVIDIA, Ising Calibration cuts calibration time from days to hours by continuously interpreting processor telemetry and adjusting control parameters without human intervention. Ising Decoding is benchmarked against pyMatching, a common reference decoder, and reports up to 2.5 times faster decoding performance and up to 3 times higher accuracy depending on the variant used.
Both models are designed to run on NVIDIA GPUs and integrate with the CUDA-Q hybrid quantum classical software stack and the NVQLink hardware interconnect, which NVIDIA positions as the bridge between classical accelerators and quantum control electronics.
How Ising compares to existing tooling
Quantum calibration and decoding have historically relied on specialised, hardware-specific software written by each lab or vendor. Ising is the first attempt to offer a general, open-weight model family trained for these tasks and shipped under a permissive license. It slots into NVIDIA’s existing open model portfolio alongside Nemotron for language and Cosmos for physical AI, extending the same distribution pattern to quantum workloads.
Compared with the classical decoders and hand-tuned calibration routines currently in use, the reported speed and accuracy gains are material rather than incremental, though real-world performance will depend on the specific quantum hardware and noise profile.
Availability and early adopters
The Ising models are available now on GitHub, Hugging Face, and build.nvidia.com, with deployment supported through NVIDIA NIM microservices for fine-tuning on specific qubit architectures. Early users include Atom Computing, Academia Sinica, Fermilab, Harvard, Infleqtion, IonQ, IQM Quantum Computers, Lawrence Berkeley National Laboratory, and the U.K. National Physical Laboratory.
For research groups and quantum hardware vendors already working inside the CUDA-Q ecosystem, Ising lowers the engineering cost of running calibration and error correction at scale. For everyone else, it sets a public baseline that other frameworks will be measured against.
The full announcement and documentation are available on the NVIDIA newsroom: https://nvidianews.nvidia.com/news/nvidia-launches-ising-the-worlds-first-open-ai-models-to-accelerate-the-path-to-useful-quantum-computers
