The Engineering Quantum Systems Group (EQuS) at the Massachusetts Institute of Technology (MIT) has announced that they have demonstrated that an AI agent based on GPT-5.6 Sol can autonomously perform the measurement and calibration of superconducting qubits.
Product LaunchesMassachusetts Institute of TechnologyEngineering Quantum Systems GroupGPT-5.6 Sol
AI Agents Autonomously Execute Measurement and Calibration of Superconducting Qubits, MIT Research Group Demonstrates
This article is a translation. Read the Japanese original
Qubit calibration is a complex process that requires hundreds to thousands of preliminary measurements and adaptive decision-making based on those results. The EQuS research group constructed a system where the AI agent is connected to experimental software, allowing it to autonomously perform a series of workflows, including selecting measurement parameters, controlling hardware, analyzing data, and determining the next experimental steps.
In the experiments, the agent autonomously completed tasks such as identifying resonator frequencies and optimizing pulse control for unmeasured 6 qubit chips. Researchers stated that because the agent can continue measurements autonomously during nighttime or other hours, it allows researchers to secure time to focus on higher-level research tasks, such as experimental design and data analysis.
On the other hand, challenges have also been reported, such as requiring more time to identify appropriate parameters or needing guidance from experienced researchers when experimental signals are weak and noisy. Furthermore, it was shown that the agent may lack the experimental intuition of an experienced researcher and might continue inappropriate investigations.
Sources
- How GPT-5.6 Sol helps run quantum computing experiments （OpenAI News、2026-09-08）
- OpenAI テクニカルケーススタディ