NVIDIA launches Ising, the first open AI models for quantum computing: calibration and error correction at 2.5x the speed of prior tools

NVIDIA launched Ising, a family of open-source AI models designed to serve as the control plane for quantum processors — the first such models publicly released. Ising addresses two foundational challenges in quantum computing: keeping processors calibrated and correcting errors in real time.

What's new

The Ising family comprises two model types:

Ising Calibration is a vision language model that interprets measurement outputs from a quantum processor and automates the continuous calibration process. Quantum processors must be recalibrated frequently because qubit fidelity drifts due to environmental noise; previously, that calibration required manual tuning cycles that took days. Ising Calibration brings that down to hours.

Ising Decoding is a pair of 3D convolutional neural network models optimized separately for speed and for accuracy. Real-time quantum error correction is one of the unsolved engineering challenges blocking practical quantum computing at scale — logical qubits require continuous syndrome decoding to remain coherent. The Ising Decoding models deliver:

• 2.5x faster processing than pyMatching, the current open-source industry standard for decoding
• 3x higher accuracy compared to pyMatching

Both variants are available on GitHub, Hugging Face, and build.nvidia.com, where they can be deployed as NVIDIA NIM microservices for direct integration into quantum computing workflows and fine-tuned on custom processor architectures.

Context

NVIDIA has been positioning itself at the intersection of classical GPU computing and emerging quantum hardware. The company's DGX Quantum systems pair NVIDIA GPUs with quantum processors over low-latency interconnects, enabling real-time classical-quantum feedback loops — the architecture that Ising Calibration and Ising Decoding are designed to run inside.

Jensen Huang, NVIDIA CEO, described the role of the models: "With Ising, AI becomes the control plane — the operating system of quantum machines — transforming fragile qubits to scalable and reliable quantum-GPU systems."

Prior to Ising, calibration and decoding for quantum processors relied on hand-tuned classical algorithms. AI-driven approaches have been explored in academic research, but a trained, open-weights model deployable via standard microservice infrastructure had not previously been released publicly.

The name Ising references the Ising model, a mathematical framework from statistical mechanics widely used to model qubit interactions and spin systems in quantum hardware design.

Why it matters

Calibration and error correction are two of the primary engineering bottlenecks preventing quantum computers from reaching commercial-scale usefulness. Cutting calibration time from days to hours is meaningful for research labs running multiple processors; making that tool open-source means quantum hardware startups and university labs gain access without requiring the engineering staff to build it from scratch.

The 3x accuracy improvement in decoding matters for threshold: quantum error correction only works when the physical error rate is below a threshold that depends on how well the decoder performs. Higher-accuracy decoding relaxes the constraints on the underlying hardware, potentially making fault-tolerant quantum computing achievable on processors that would otherwise fall short.

NVIDIA's move to open-source these models signals that the company sees AI-driven quantum control as an emerging ecosystem it wants to be central to — analogous to how CUDA established NVIDIA's position in GPU computing by giving researchers a free toolkit to build on.