NVIDIA Releases World's First Open Quantum AI Models — Error Correction Gets 3× More Accurate
NVIDIA launched Ising, the first family of open AI models designed specifically for quantum computing — a 35B parameter VLM for processor calibration and a 3D CNN decoder for error correction that's 2.5× faster and 3× more accurate than existing tools.
Original sourceNVIDIA released Ising on April 14, 2026 — a family of open AI models that address quantum computing's two most persistent engineering obstacles: calibrating quantum processors and correcting quantum errors fast enough for real-world applications.
The family has two distinct models. **Ising Calibration** is a 35-billion-parameter vision-language model trained on multi-modality qubit data. It automates quantum processor calibration — a process that previously took days of painstaking manual tuning — down to hours, and outperforms GPT 5.4, Claude Opus 4.6, and Gemini 3.1 Pro on the QCalEval benchmark for quantum calibration tasks. **Ising Decoding** is a 3D CNN-based framework for real-time quantum error correction that delivers up to 2.5× faster performance and 3× higher accuracy compared to pyMatching, the current industry standard.
Both models are available open-weight on GitHub and Hugging Face, integrate with NVIDIA's CUDA-Q software platform, and deploy via NVIDIA NIM microservices. Early adopters include Harvard, Fermi National Accelerator Lab, Lawrence Berkeley National Lab, and the UK's National Physical Laboratory.
Jensen Huang framed it simply at the announcement: "AI is essential to making quantum computing practical... AI becomes the control plane — the operating system of quantum machines." With the quantum computing market projected to exceed $11 billion by 2030, the release of open models for core quantum infrastructure is a significant forcing function for the whole industry. For quantum hardware companies, Ising eliminates one of the most expensive and time-consuming steps in building useful systems.
The open-weight release is particularly notable. NVIDIA is betting that accelerating the quantum ecosystem broadly — rather than locking away the models — creates more value for its CUDA-Q and NVQLink hardware stack than any licensing approach could. It's the same playbook that made CUDA the default GPU compute platform.
Panel Takes
The Builder
Developer Perspective
“Open weights on Hugging Face means quantum researchers at universities and startups can actually use these without an enterprise contract. The calibration automation alone — collapsing days to hours — is a practical breakthrough for labs that spend more time calibrating than computing.”
The Skeptic
Reality Check
“Ising requires CUDA-Q and ideally NVQLink hardware to get full performance — this is open models with NVIDIA-shaped lock-in baked in. The QCalEval benchmark is also NVIDIA's own benchmark, which makes beating GPT and Claude on it less impressive than it sounds.”
The Futurist
Big Picture
“AI as the operating system of quantum machines is one of those framings that sounds like marketing until you realize it's architecturally accurate. The bet here is that whoever controls the AI control plane for quantum hardware ends up owning the quantum stack — and NVIDIA is making that play before anyone else.”