OpenAI put GPT-5.6 Sol on a real six-qubit chip at MIT and let it calibrate the hardware unsupervised
In a 2026-09-08 post, OpenAI describes MIT Engineering Quantum Systems graduate student Beatriz Yankelevich harnessing GPT-5.6 Sol through Codex to lab software controlling a six-qubit superconducting chip in a dilution refrigerator. The agent selected experimental settings, drove the hardware, analysed data and calibrated qubit frequencies, control pulses and information-retention times autonomously. The honest limit is stated: it performed well on clear signals but needed expert guidance when results were weak, noisy or physically ambiguous. Corroborated by The Quantum Insider on 2026-09-09. The transferable point for builders is that the agent's job was routine measurement loops, not experiment design.
Source
↳ Follow the thread