NVIDIA Ising
AI model · Science / biology model by Nvidia
NVIDIA Ising is a family of open AI models that accelerate quantum computing by improving processor calibration and error correction. These models offer significant speed and accuracy improvements over traditional methods, empowering researchers and enterprises to build scalable quantum systems.
- Modalities
- Image, Text → Text
- Released
- 2026-04-14
- Release Date
- April 14, 2026
- License
- NVIDIA Open Model License, Apache 2.0
- Primary Use Case
- Quantum Computing (Calibration & Error Correction)
- Performance Improvement (Decoding)
- Up to 2.5x faster, 3x more accurate than pyMatching
- World's first open AI models for quantum computing
- Breakthrough performance in quantum calibration and error correction
- Up to 2.5x faster and 3x more accurate than traditional methods
- Automates QPU calibration, reducing bring-up time from days to hours
- Supports SurfaceCode and ColorCode with flexible noise model training
Overview
NVIDIA Ising is a family of open-source AI models designed to accelerate the development of practical quantum computers. It addresses critical challenges in quantum computing, specifically quantum processor calibration and quantum error correction. By providing state-of-the-art AI tools, Ising empowers researchers and enterprises to build scalable, high-performance quantum systems. The models automate complex tasks, significantly reducing the time and increasing the accuracy compared to traditional methods, making quantum computing more accessible and reliable.
Quantum Computing Tasks
- Automated QPU Calibration: Automates continuous calibration of quantum processors, reducing bring-up time from days to hours.
- Real-time Quantum Error Correction Decoding: Performs real-time decoding for quantum error correction using optimized 3D CNN models.
- Faster, More Accurate Decoding: Accelerates the decoding process for quantum error correction, achieving up to 2.5x faster performance and 3x higher accuracy than pyMatching.
- Flexible Noise Model Support: Supports SurfaceCode and ColorCode with off-the-shelf noise models, and includes a new training framework for any noise model.
- Agentic Workflow Automation for Calibration: Enables AI agents to automate continuous calibration and retune workflows.
- Scaling to Fault Tolerance: Provides AI tools to scale quantum devices to fault tolerance.
Use cases
- Automating quantum processor calibration to reduce bring-up time.
- Performing real-time decoding for quantum error correction with high accuracy and speed.
- Enabling researchers to build scalable and reliable quantum-GPU systems.
- Supporting the development of fault-tolerant quantum computers.
- Accelerating the path to useful quantum applications.
Specifications
- Modalities
- Image, Text → Text
- Parameters
- 35B (Ising-Calibration-1-35B-A3B), 31B (Ising-Calibration-1.5-31B-BF16), 1.8M (Ising-Decoder-SurfaceCode-1-Accurate), 0.91M (Ising-Decoder-SurfaceCode-1-Fast), 2.94M (Ising-Decoder-ColorCode-1-Fast)
- Licence
- NVIDIA Open Model License, Apache License 2.0 (for Qwen3.5-35B-A3B base model)
- Released
- 2026-04-14
- Release Date
- April 14, 2026
- Model Variants
- Ising-Calibration-1-35B-A3B, Ising-Calibration-1.5-31B-BF16, Ising-Calibration-1.5-31B-NVFP4, Ising-Decoder-SurfaceCode-1-Accurate, Ising-Decoder-SurfaceCode-1-Fast, Ising-Decoder-ColorCode-1-Fast
- License
- NVIDIA Open Model License, Apache 2.0
- Primary Use Case
- Quantum Computing (Calibration & Error Correction)
- Performance Improvement (Decoding)
- Up to 2.5x faster, 3x more accurate than pyMatching
- Calibration Automation
- Reduces bring-up time from days to hours
How it works
- Access Models & APIs: Access pre-trained models and tooling on Hugging Face or via NVIDIA NIM API endpoints.
- Integrate into Workflows: Integrate Ising models into quantum computing workflows using provided frameworks and guidelines.
- Customize and Deploy: Retrain, fine-tune, and deploy models using PyTorch and CUDA-Q for specific quantum computing tasks.
FAQ
What is NVIDIA Ising?
NVIDIA Ising is a family of open-source AI models designed to accelerate the development of practical quantum computers. It addresses critical challenges in quantum computing, specifically quantum processor calibration and quantum error correction. By providing state-of-the-art AI tools, Ising empowers researchers and enterprises to build scalable, high-performance quantum systems.
How can I access and use NVIDIA Ising models?
NVIDIA Ising models are available on Hugging Face and can be accessed via NVIDIA NIM API endpoints. They can also be downloaded and deployed using frameworks like PyTorch and CUDA-Q.
What are the primary use cases for NVIDIA Ising?
NVIDIA Ising models are designed for quantum computing tasks such as automating quantum processor calibration and performing real-time decoding for quantum error correction.
What are the performance benefits of using NVIDIA Ising?
NVIDIA Ising offers significant performance improvements, with decoding processes up to 2.5x faster and 3x more accurate than traditional methods like pyMatching. Calibration automation can reduce bring-up time from days to hours.
Does NVIDIA Ising support different noise models?
Yes, NVIDIA Ising supports various noise models for quantum error correction and includes a new training framework to support any noise model through PyTorch and CUDA-Q.
What are the different model variants and their parameter counts?
The Ising model family includes variants like Ising-Calibration-1-35B-A3B (35B parameters), Ising-Calibration-1.5-31B-BF16 (31B parameters), and various decoder models with parameter counts ranging from ~0.91M to 2.94M.