NVIDIA Releases Open-Source AI Model for Quantum Computer Tuning

NVIDIA has released Ising Calibration 1.5, an open-source vision language model designed to automate the calibration of quantum processing units (QPUs) by interpreting their diagnostic outputs and determining necessary adjustments.

The model represents a significant step toward fully automated quantum computer operation. Unlike traditional calibration approaches that require extensive manual intervention, Ising Calibration 1.5 can analyze unfamiliar diagnostic results without prior training examples—a capability crucial for maintaining quantum systems as they drift out of tune during operation.

Technical capabilities and deployment options

The 31-billion-parameter model works across multiple qubit modalities, including superconducting qubits, quantum dots, ions, neutral atoms, and electrons on helium. Training data came from partner contributions spanning these diverse quantum computing architectures.

Ising Calibration 1.5 delivers an 86.68% improvement over its predecessor when analyzing results in the context of related experiments—a mode called in-context learning. The model is also 11.4% smaller at BF16 precision compared to the previous version, making it easier to deploy in laboratory environments.

For the first time, NVIDIA is offering the model in an NVFP4-quantized version that runs on a single GPU or an NVIDIA DGX Spark system. This quantized variant enables deployment on consumer-grade hardware with minimal accuracy loss, bringing enterprise-class quantum calibration capabilities to smaller research facilities.

Benchmark performance

NVIDIA evaluated the model using QCalEval, a benchmark that measures how well AI systems can interpret experimental results, classify outcomes, assess fit quality, and recommend next steps for quantum calibration tasks.

According to the evaluation, Ising Calibration 1.5 outperforms all comparable open models and remains competitive with leading closed models like Fable 5 and GPT 5.6 Sol, despite being fully open-source.

Why it matters

Quantum computers require constant recalibration as environmental factors and hardware drift degrade performance. Manual calibration by specialized engineers is time-consuming and expensive, creating a bottleneck for quantum computing scalability. An AI system that can autonomously interpret diagnostic plots and recommend tuning adjustments could dramatically reduce operational costs and enable quantum systems to self-correct between experiments. The open-source release also allows quantum hardware developers to customize the model for their specific architectures rather than depending on proprietary solutions.

Open access and integration

The model weights, training data, benchmarks, and recipes are available on Hugging Face under the OpenMDW License from the Linux Foundation. NVIDIA also provides the model as an NVIDIA NIM and through NVIDIA Build for cloud deployment.

A ready-to-use agent blueprint is available on GitHub at NVIDIA/Quantum-Calibration-Agent-Blueprint, offering integration support for both the Ising model and large cloud-based language model APIs.

These details were first reported by NVIDIA in a developer blog post announcing the release.