QCalEval: Benchmarking Vision-Language Models for Quantum Calibration Plot Understanding

TL;DR

This paper introduces QCalEval, a benchmark for vision-language models to interpret quantum calibration plots, revealing strengths and limitations of current models in this specialized domain.

Contribution

It presents the first systematic evaluation of VLMs on quantum calibration plots, including a new benchmark and analysis of model performance in zero-shot and in-context settings.

Findings

Best zero-shot model scores 72.3 on average.

In-context learning degrades performance for many models.

Supervised fine-tuning improves zero-shot scores but not in-context learning.

Abstract

Quantum computing calibration depends on interpreting experimental data, and calibration plots provide the most universal human-readable representation for this task, yet no systematic evaluation exists of how well vision-language models (VLMs) interpret them. We introduce QCalEval, the first VLM benchmark for quantum calibration plots: 243 samples across 87 scenario types from 22 experiment families, spanning superconducting qubits and neutral atoms, evaluated on six question types in both zero-shot and in-context learning settings. The best general-purpose zero-shot model reaches a mean score of 72.3, and many open-weight models degrade under multi-image in-context learning, whereas frontier closed models improve substantially. A supervised fine-tuning ablation at the 9-billion-parameter scale shows that SFT improves zero-shot performance but cannot close the multimodal in-context learning gap. As a reference case study, we release NVIDIA Ising Calibration 1, an open-weight model based on Qwen3.5-35B-A3B that reaches 74.7 zero-shot average score.