NVIDIA has released NVIDIA Ising Calibration 1.5 as an open-source AI model designed to automate parts of the quantum computer calibration process. According to the company, the model can analyze diagnostic data from quantum processing units, or QPUs, and determine device tuning plans on its own.
NVIDIA described Ising Calibration 1.5 as a vision-language model, or VLM, built specifically for quantum calibration work. It is designed to understand experimental data generated by quantum chips. The company said the model can carry out zero-shot analysis when historical examples are not available, and it can also use in-context learning, or ICL, with related experimental samples to help quantum devices maintain and improve operating conditions.
Benchmark performance on QCalEval
In the QCalEval quantum calibration benchmark, NVIDIA said Ising Calibration 1.5 outperformed open-source models of similar size by about 10% on average in zero-shot reasoning. When related experiment cases were provided for in-context learning, performance improved by about 86.5% from the previous generation. The company added that the model also surpassed several open-source models and came close to the level of top closed-source large models.
31 billion parameters and broader deployment options
The model has 31 billion parameters and supports deployment on NVIDIA Grace Blackwell and Vera Rubin data center GPUs. NVIDIA also released an NVFP4 quantized version, which the company said can run on a single consumer-grade GPU or on NVIDIA DGX Spark. That setup is intended to lower the deployment threshold for quantum laboratories.
Training data spans multiple qubit architectures
NVIDIA said the model was trained on data drawn from multiple qubit architectures, including superconducting qubits, quantum dots, ions, neutral atoms, and electrons on helium. The company said this allows the model to provide calibration support across different types of quantum computing hardware.
The report noted that automated calibration is widely seen in the industry as one of the key bottlenecks in scaling quantum computing. NVIDIA’s release of an AI-driven calibration tool points to AI models moving beyond traditional computing tasks and into the hardware control layer for quantum systems.