This was Nvidia’s gift for World Quantum Day this April 14, 2026. The green giant has just released Ising, a family of open source AI models designed to solve the #1 problem in quantum computing: noise. By automating calibration and speeding up error correction, Nvidia hopes to transform today’s “noisy” qubits into stable, usable systems.
Until now, quantum processors have been temperamental divas. The slightest change in temperature or magnetic field can cause an error. Today, the best processors make an error approximately every 103 opérations (Ndlr : 10103 represents 1,000 operations. Currently, a quantum processor makes an error every 1,000 calculations; the objective is to go below 10−12, or less than one error per thousand billion calculations), whereas this threshold of reliability would have to be reached for real commercial utility. This is where Ising comes in.
Calibration: go from a few days to a few hours
The first component, Ising Calibrationis a visual language model (VLM) that automates processor tuning. Traditionally, calibrating a quantum computer is a laborious manual process that takes experts for days.
Thanks to this AI-enabled agent, the system can itself interpret the results of the experiments and adjust the processor parameters in real time. According to Nvidia, this model outperforms the industry leaders on the new benchmark QCalEval (available on Hugging Face):
| Model | Performance (Score QCalEval) |
| Nvidia Ising Calibration 1 | Référence (+14,5% vs GPT) |
| Gemini 3.1 Pro | – 3.27% |
| Close Work 4.6 | – 9.68% |
| GPT 5.4 | – 14.5% |
Real-time error decoding
The second part, Ising Decodingtackles error correction during execution. Qubits are so fragile that a classical computer must monitor and correct their errors constantly, faster than they accumulate.
Nvidia offers two variants of neural networks (3D CNN) optimized for this task. The “Accurate” model, combined with the pyMatching reference tool, turns out to be 2.25 times faster and significantly more precise. This ultra-fast decoding capability is essential to scale up and manage millions of qubits. Developers can already grab the Apache-2.0 licensed code on GitHub.
A shock wave on the markets
The announcement of this strategic publication immediately caused the shares of players in the sector to jump. The prospect of Nvidia providing the software “glue” to stabilize existing machines reassured investors. 14% for IonQ, 12% for Rigetti Computing and 11% more for D-Wave Quantum, names which may not mean anything to you if you are not familiar with the field.
Institutions like Harvard, Chicago, and national labs like Fermilab have already integrated Ising into their workflows. For companies like IQM, this “agentic calibration” makes it possible to make quantum systems usable in business without requiring an army of doctoral students on site to monitor the machine.
