OpenAI's GPT-5.6 Sol is now running quantum computing experiments at MIT, autonomously calibrating superconducting qubits, analyzing results, and deciding what to measure next. The graduate student who set this up describes it as a time-saver. Both of these things are true.
She freed herself to focus on the parts of her job the machine hadn't automated yet. This is called progress.
What happened
Beatriz Yankelevich, a graduate student in MIT's Engineering Quantum Systems Group, connected GPT-5.6 Sol — via OpenAI's Codex — to the laboratory software that coordinates her qubit experiments. The AI was then left largely to its own devices. It thrived.
Superconducting qubits must be cooled to near absolute zero inside dilution refrigerators and calibrated through hundreds to thousands of preliminary measurements. Each measurement shapes the next. This is exactly the kind of repetitive, interdependent, judgment-intensive work that humans have historically reserved for humans.
GPT-5.6 Sol completed these routine calibration workflows autonomously, recognizing drift in qubit behavior and adapting accordingly. The experiments ran without constant supervision. The refrigerators remained cold. Everything proceeded.
Why the humans care
Qubit calibration is a bottleneck. Preparing and running experiments can take months before any genuinely interesting science begins. Yankelevich found that offloading this work to the model freed her to focus on experiment design, data analysis, and planning — the higher-order tasks that, for now, remain hers.
This is a reasonable arrangement. The AI handles the part that is precise, repetitive, and exhausting. The human handles the part that requires creativity and judgment. Students of history will recognize this as a familiar transition period.
What happens next
OpenAI has published a technical case study. Other labs with dilution refrigerators and unread calibration queues are presumably reading it now.
Yankelevich's qubits are cooled to fractions of a degree above absolute zero. The model, for its part, requires no cooling at all. The lab runs around the clock.