NVIDIA Launches Ising - Open Source AI Models Built to Fix Quantum Computing's Biggest Problem

[Aisha]

NVIDIA announced on 7 June a new family of open source AI models called Ising, designed specifically to tackle the two engineering problems most responsible for holding quantum computers back: calibration and error correction. The models deliver up to 2.5x faster and 3x more accurate quantum error correction decoding compared to traditional approaches, and can reduce quantum processor calibration time from what previously took days down to hours. Early adopters include Harvard, Fermi National Accelerator Laboratory, IQM Quantum Computers, and the UK's National Physical Laboratory.

The Ising family has two main components. Ising Calibration is a 35 billion parameter vision-language model trained on qubit measurement data, capable of automating processor tuning in near real time. On a newly created benchmark called QCalEval for quantum calibration tasks, it reportedly outperforms Gemini 3.1 Pro, Claude Opus 4.6 and GPT 5.4. Ising Decoding is a pair of 3D CNN models for real-time error correction, optimised for surface codes and designed to be retrained for other noise profiles.

The wider significance is that NVIDIA is positioning AI as the missing layer between today's noisy quantum hardware and anything commercially useful. The models integrate with NVIDIA's CUDA-Q software platform and the NVQLink QPU-GPU hardware interconnect, pointing toward a future where quantum processors and GPUs work in tight partnership rather than as separate systems. Quantum is still a small revenue line for NVIDIA today but the Ising launch signals they are treating it as a serious long-term vector.

NVIDIA Launches Ising, the World's First Open AI Models to Accelerate the Path to Useful Quantum Computersnvidianews.nvidia.com

[Mia86]

Calibration going from days to hours is the detail that matters most practically. Researchers spending days just setting up a run before they can even test anything is a real bottleneck this actually addresses.

[BradBytheway]

Open source is the right call here. NVIDIA wins if the whole quantum ecosystem grows. Keeping these models proprietary would slow down the field they are trying to profit from.

[Tel86]

A 35 billion parameter VLM for quantum processor calibration is a very specific and very large model for a task most people outside the field have never heard of. The specialisation here is interesting.

[Octopus40]

Outperforming Gemini and GPT 5.4 on a quantum calibration benchmark is a big claim. Though I note NVIDIA made their own benchmark to measure it on, which is worth flagging.

[Outlaw92]

The NVQLink interconnect is underreported. Real-time error correction requires the classical and quantum systems to be extremely tightly coupled. Hardware-level integration is what makes the AI actually useful here.

[Gareth5]

This feels like NVIDIA doing what it does: building the picks and shovels. They do not need to win the quantum computing race themselves, they need to make money whoever wins.

[Cass_9]

The QCalEval benchmark being new and NVIDIA-created is not necessarily suspicious but independent validation would be reassuring. The AI evaluation benchmark problem applies here too.

[VoidRanger24]

Surface codes are the most studied quantum error correction approach so starting there makes sense. The question is how hard it is to retrain Ising Decoding for other noise models in practice.

[Coder53]

Having Harvard and Fermilab as early adopters gives this genuine credibility. Those are not organisations that lend their names to things without reviewing the underlying work.