NVIDIA · 2026-04-14 · notable
NVIDIA Ising — Open AI Models for Quantum Processor Calibration and Error Correction
NVIDIA released Ising, the first open AI models for quantum computing: a 35B MoE VLM that automates quantum processor calibration and 3D CNN decoders for real-time error correction, 2.5× faster and 3× more accurate than pyMatching.
NVIDIA's Ising family applies AI to two core quantum computing bottlenecks: calibrating noisy qubits and decoding quantum errors in real time.
Key specs
| License | NVIDIA Open Model License / Apache 2.0 (framework) |
|---|---|
| Calibration params | 35B total / 3B active (MoE) |
| Qcal eval score | 74.7% |
| Decoder vs py matching (speed) | 2.5× |
| Decoder vs py matching (accuracy) | 3× |
What is it?
NVIDIA Ising is a family of open AI models designed to help quantum hardware teams and researchers run fault-tolerant quantum processors. It has two parts: Ising Calibration, a 35B-parameter MoE vision-language model that interprets experimental qubit calibration plots and runs agentic calibration workflows; and Ising Decoding, lightweight 3D CNN models (~1–2M parameters) for real-time quantum error correction. Model weights are on HuggingFace; training frameworks are on GitHub under Apache 2.0.
How does it work?
Ising Calibration is built on Qwen3.5-35B-A3B (256 experts, 8 active), fine-tuned on 72.5K calibration experiment images with structured technical text. It answers six question types about each plot — technical description, experimental conclusion, fit quality, parameter extraction, and success classification — scoring 74.7% on the newly released QCalEval benchmark, outperforming Gemini 3.1 Pro, Claude Opus 4.6, and GPT 5.4. Ising Decoding uses 3D CNNs to predict quantum error corrections from syndrome measurements across space and time, with FP8 quantization for low-latency inference. Both integrate with NVIDIA's CUDA-Q quantum software platform.
Why does it matter?
Manual quantum processor calibration typically takes days per device; Ising Calibration automates it using the same experimental plots a physicist would read. The decoding models need 10× less training data than alternatives and run 2.5× faster with 3× better accuracy than pyMatching, the open-source baseline most quantum research groups use. Major labs including Fermilab, Harvard, and the UK National Physical Laboratory are adopting the models.
Who is it for?
Quantum hardware teams and academic research groups working on superconducting qubits or neutral atom processors.
Try it
huggingface.co/nvidia/Ising-Calibration-1-35B-A3B — runs on 2× NVIDIA L40S (48 GB) or 1× H100 (80 GB)