NVIDIA has launched Ising, described as the world's first family of open AI models for building fault-tolerant quantum processors. The family includes two model domains: Ising Calibration, a vision-language model (VLM) that automates quantum processor calibration using agentic workflows, and Ising Decoding, consisting of 3D CNN models for real-time quantum error correction decoding. Ising Calibration outperforms competing models on the newly created QCalEval benchmark. Ising Decoding offers Fast and Accurate variants with measurable speedups and logical error rate improvements over baseline PyMatching. All model weights, training frameworks, datasets, and deployment recipes are open and available on HuggingFace and GitHub under Apache 2.0, enabling QPU builders to fine-tune models for their specific hardware and noise characteristics.
Table of contents
NVIDIA Ising CalibrationNVIDIA Ising DecodingStart building with NVIDIA Ising open resourcesGet startedSort: