NVIDIA Launches Ising: First Open Quantum AI Model Family

NVIDIA has officially launched NVIDIA Ising, the world’s first family of open quantum AI models designed to bridge the gap between experimental quantum research and practical, real-world applications. By leveraging artificial intelligence to address the critical engineering bottlenecks of quantum calibration and error correction, the Ising model family aims to transform how researchers and enterprises manage the sensitivity of quantum processors.

Automating Quantum Calibration

Quantum computers are notoriously sensitive to environmental noise, which causes qubits to accumulate errors rapidly during computation. Historically, maintaining these systems has required slow, manual intervention to ensure hardware is properly tuned. NVIDIA Ising addresses this through Ising Calibration, a vision language model architecture that functions as an autonomous AI agent.
By continuously monitoring diagnostic readouts from quantum hardware, Ising Calibration can autonomously adjust the system to maintain optimal performance. This shift from manual oversight to AI-driven automation significantly accelerates the development process, reducing the time required for continuous calibration from days to hours.

Advancing Real-Time Error Correction

Beyond calibration, the Ising family introduces Ising Decoding, which utilizes 3D convolutional neural networks (3D CNNs) to perform real-time quantum error correction. These models are designed to infer the correct state of a system by analyzing noisy observations, a process conceptually similar to signal processing.
NVIDIA offers two variants of the Ising Decoding model, allowing researchers to prioritize either speed or accuracy. Performance benchmarks indicate that these models are up to 2.5 times faster and 3 times more accurate than pyMatching, which currently serves as the industry standard for open-source error correction.

Integration and Ecosystem Adoption

The Ising model family is designed to integrate seamlessly into NVIDIA’s existing quantum-classical ecosystem. It complements the CUDA-Q software platform, which provides a programming model for hybrid workflows, and utilizes the NVQLink QPU-GPU hardware interconnect to facilitate the low-latency communication necessary for real-time error correction.
The adoption of these models is already widespread, spanning national laboratories, Ivy League universities, and commercial hardware companies. Organizations such as Fermi National Accelerator Laboratory, Harvard University, IonQ, IQM Quantum Computers, and Sandia National Laboratories are among the many institutions deploying Ising Calibration and Ising Decoding. The models are currently available on GitHub, Hugging Face, and build.nvidia.com, with support for fine-tuning via NVIDIA NIM microservices.