NVIDIA Ising Introduces AI-Powered Workflows to Build Fault-Tolerant Quantum Systems

2026-04-14 · Source: NVIDIA Technical Blog · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Emerging Technologies & Innovation · Depth: Expert, medium

Summary

NVIDIA Ising is the first open AI model family designed to build quantum processors, addressing the critical challenge of qubit noise in quantum computing. It launches with two model domains: Ising Calibration and Ising Decoding. Ising Calibration is a 35B parameter vision-language model (VLM) that automates quantum processing unit (QPU) calibration, outperforming models like Gemini 3.1 Pro by 3.27% on the new QCalEval benchmark. Ising Decoding comprises two 3D CNN models for real-time quantum error correction decoding, offering both "Fast" (912,000 parameters) and "Accurate" (1.79 million parameters) versions. These models, available on HuggingFace, improve logical error rates and latency, with the Accurate pre-decoder plus PyMatching being 2.25x faster and 1.53x more accurate than PyMatching alone for d=13 at p=0.003. NVIDIA provides open base models, training frameworks, and deployment workflows to enable customization and scaling to millions of qubits.

Key takeaway

For quantum computing engineers and researchers focused on improving QPU reliability, NVIDIA Ising offers a critical open-source toolkit. You should explore integrating the Ising Calibration VLM for automated QPU tuning and leverage the Ising Decoding 3D CNNs to enhance real-time error correction. This approach can significantly reduce qubit error rates and accelerate the path toward fault-tolerant quantum systems, allowing you to tailor models to your specific hardware and noise characteristics.

Key insights

NVIDIA Ising provides open AI models and tools to mitigate qubit noise and scale quantum computing.

Principles

• AI is key to reducing quantum error rates.
• Open models accelerate quantum hardware development.
• Calibration and decoding are critical for QPU performance.

Method

NVIDIA Ising uses VLMs for agentic QPU calibration and 3D CNNs for real-time quantum error correction decoding, supported by a training framework for synthetic data generation and fine-tuning.

In practice

• Use Ising Calibration 1 for automated QPU tuning.
• Deploy Ising Decoding models for real-time error correction.
• Fine-tune models with cuStabilizer for specific QPU noise.

Topics

• NVIDIA Ising
• Quantum Error Correction
• Qubit Calibration
• Vision-Language Models
• 3D CNN Decoders

Code references

• NVIDIA/Quantum-Calibration-Agent-Blueprint
• oscarhiggott/PyMatching
• NVIDIA/Ising-Decoding

Best for: AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• NVIDIA's Quantum Day | here's a glimpse into the future... — NVIDIA's open Ising models accelerate quantum computing by integrating AI with GPU supercomputing for critical calibration and error correction tasks.
• Introducing NVIDIA Ising — NVIDIA's Ising AI models automate quantum processor calibration and error correction, accelerating quantum computing development.
• Advanced MXFP4 Quantization: Combining Fine-Tuned Rotations with SmoothQuant for Near-Lossless Compression — Combining fine-tuned online rotations and SmoothQuant scales significantly improves MXFP4 quantization accuracy for language models.
• AI for Quantum: NVIDIA Ising Accelerates Useful Quantum Computing — NVIDIA Ising AI models accelerate quantum computing by automating calibration and error correction.
• Quantum computer breakthrough tracks qubit fluctuations in real time — Real-time adaptive measurement of qubit relaxation rates is crucial for stabilizing and scaling quantum processors.
• AQ4SViT: An Automated Quantization Framework with Search Gating Policy for Compressing Spiking Vision Transformers — Automated quantization for SViTs uses performance proxies to quickly find optimal settings with accuracy-memory trade-offs.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by NVIDIA Technical Blog.