NVIDIA announced the world's first family of open-source quantum AI models called NVIDIA Ising in mid-April 2026, designed to accelerate the development of quantum processors capable of running useful applications. The models address two critical bottlenecks in quantum computing: continuous calibration and quantum error correction, delivering up to 2.5× faster and 3× more accurate results than current industry standards.
Two Model Families Target Different Quantum Computing Challenges
NVIDIA Ising comprises two distinct model domains, each addressing a specific technical challenge in quantum processor development:
Ising Calibration: A 35 billion parameter Vision Language Model fine-tuned to infer calibration actions from quantum processor experimental data. This model enables AI agents to automate continuous calibration, reducing the time required from days to hours. The model can rapidly interpret and react to measurements from quantum processors, streamlining a previously time-intensive manual process.
Ising Decoding: Two variants of a 3D convolutional neural network model with either 0.9 million or 1.8 million parameters, optimized for speed or accuracy respectively. These models perform real-time decoding for quantum error correction, delivering up to 2.5× faster and 3× more accurate results than pyMatching, the current open-source industry standard.
Full Open-Source Release Under Permissive Licensing
NVIDIA released Ising under the Apache 2.0 license, making it fully open-source and permissively licensed for commercial use. Model weights are available on Hugging Face and build.nvidia.com, while training frameworks are accessible on GitHub. The release includes models working with a depolarizing noise model for surface codes of any distance, plus a new training framework supporting any noise model through PyTorch and CUDA-Q.
This represents NVIDIA's strategic bet that AI will become essential infrastructure for making quantum computing practical, similar to how AI has become critical for optimizing classical computing systems.
Addressing Real-World Quantum Computing Bottlenecks
The models target two of the most time-consuming and error-prone aspects of quantum processor development. Continuous calibration has traditionally required days of manual tuning, while error correction accuracy has limited the practical application of quantum systems. By automating calibration and improving error correction performance, Ising addresses fundamental obstacles preventing fault-tolerant quantum computing at scale.
The release comes as quantum computing transitions from research to practical applications, providing infrastructure to help bridge this gap. The performance improvements over existing tools—2.5× speed and 3× accuracy compared to pyMatching—represent meaningful advances that could accelerate quantum computing development timelines across the industry.
Key Takeaways
- NVIDIA announced NVIDIA Ising in mid-April 2026 as the world's first family of open-source quantum AI models, targeting quantum processor calibration and error correction
- Ising Calibration uses a 35 billion parameter Vision Language Model to reduce quantum processor calibration time from days to hours through automated interpretation of experimental data
- Ising Decoding models (0.9M or 1.8M parameters) deliver 2.5× faster and 3× more accurate quantum error correction compared to pyMatching, the current industry standard
- Model weights are available on Hugging Face and build.nvidia.com, with training frameworks on GitHub under the permissive Apache 2.0 license
- The release addresses critical bottlenecks preventing fault-tolerant quantum computing at scale, positioning AI as essential infrastructure for practical quantum computing