NVIDIA Launches Ising — Open-Source Quantum AI Models (April 2026)

NVIDIA on April 14, 2026 — World Quantum Day — released NVIDIA Ising, the world's first family of open-source AI models purpose-built for quantum computing, shipping under Apache 2.0 on GitHub and Hugging Face. The release ships two model families: a 35-billion-parameter vision-language model for qubit calibration and a 3D convolutional neural network that performs surface-code quantum error correction 2.5× faster and 3× more accurate than pyMatching, the open decoder most quantum labs use today.

What Happened

NVIDIA introduced Ising as the first production-ready open AI toolkit for the two hardest bottlenecks in modern quantum computing: keeping qubits tuned (calibration) and correcting the errors they produce at runtime (decoding). The announcement came on April 14, 2026 via the NVIDIA Newsroom and a companion NVIDIA Developer blog post.

"AI is essential to making quantum computing practical," said NVIDIA founder and CEO Jensen Huang. "With Ising, AI becomes the control plane — the operating system of quantum machines — transforming fragile qubits to scalable and reliable quantum-GPU systems." Ising is designed to complement NVIDIA's existing CUDA-Q hybrid quantum-classical software platform and the NVQLink QPU-GPU interconnect, so the same servers that train frontier LLMs can now also steer a cryogenic quantum processor in real time.

Key Details

• Ising Calibration — a 35B-parameter vision-language model fine-tuned on multi-modal qubit measurements. It reads experimental traces from a QPU and infers the control-parameter adjustments needed to re-tune it, cutting calibration time from days to hours when paired with an agent. NVIDIA reports it outperforms Gemini 3.1 Pro, Claude Opus 4.6, and GPT-5.4 on the newly introduced QCalEval benchmark.
• Ising Decoding — two variants of a 3D CNN (0.9M and 1.8M parameters) that pre-decode surface-code quantum error correction at up to 2.5× the speed and 3× the accuracy of pyMatching, the open-source decoder standard used across the field. NVIDIA says Ising Decoding needs 10× less training data to reach that accuracy.
• License and distribution — both model families ship under the Apache 2.0 license, with weights and inference code on GitHub and Hugging Face; integration points land in NVIDIA's CUDA-Q stack and the NVQLink hardware interconnect.
• Immediate adopters — Ising Calibration is already in use at Atom Computing, IonQ, IQM Quantum Computers, Infleqtion, EeroQ, Conductor Quantum, Fermi National Accelerator Laboratory, Harvard, Academia Sinica, Q-CTRL, Lawrence Berkeley National Lab, and the U.K. National Physical Laboratory. Ising Decoding is deployed at Cornell, Sandia National Laboratories, EdenCode, Quantum Elements, and others.

What Developers and Users Are Saying

Reaction across Tom's Hardware, HPCwire, and The Quantum Insider has been notably positive. Quantum-hardware analysts frame Ising as the first credible attempt to turn AI into a deployable quantum control stack rather than a research demo — especially because the open license lets hardware vendors integrate the decoder without paying a per-QPU royalty.

The market moved quickly: quantum-pure-play stocks rallied in the days following the announcement. Per CNBC coverage on April 16, IonQ and D-Wave shares rose more than 50% on the week, while Rigetti and Quantum Computing Inc. both gained over 30%. The more skeptical thread in developer discussions: calibration and decoding are still only two of many unsolved problems on the path to fault-tolerant quantum computing, and an open model family does not by itself extend qubit coherence times or reduce gate error rates.

What This Means for Developers

For quantum software developers, Ising removes two of the heaviest engineering lifts from day-one experimentation: you can now bolt an AI-based error-correction decoder onto any surface-code workload and re-use NVIDIA's calibration VLM instead of writing your own characterization pipeline. Both ship as drop-in components to the CUDA-Q runtime.

For AI infrastructure engineers, Ising is another signal that NVIDIA intends to own every classical-to-physical control loop — GPU clusters are now the operating system for LLMs, robotics, autonomous vehicles, and, as of this week, quantum processors. Expect NVQLink-enabled server SKUs and a new class of "hybrid quantum-GPU" reference architectures to follow later in 2026.

For everyone else, the practical test is still quantum advantage on a useful workload. Ising does not deliver that today; it removes one of the biggest classical engineering barriers on the path to it.

What's Next

NVIDIA says Ising will evolve alongside CUDA-Q and NVQLink, with additional model variants planned for higher-dimensional error-correction codes and for hardware architectures beyond surface codes (LDPC and color codes, for example). Weights, inference code, and benchmark harnesses are available today on GitHub and Hugging Face; deeper integration hooks into CUDA-Q are documented on the NVIDIA Developer page for Ising.

Sources

• NVIDIA Newsroom — official launch announcement, April 14, 2026
• NVIDIA Developer Blog — technical deep-dive on calibration and decoding
• Tom's Hardware — independent coverage of benchmark claims
• HPCwire — HPC community analysis
• The Quantum Insider — quantum industry reporting
• CNBC — market reaction and quantum-stock rally
• MarkTechPost — developer-focused recap