OpenAI's Codex runs qubit calibrations at MIT
OpenAI says a Codex agent running GPT‑5.6 Sol autonomously calibrated superconducting qubits in an MIT lab, an early test of whether AI agents can operate physical experiments without constant supervision.
OpenAI has published a case study describing how a graduate researcher at MIT wired its Codex agent, running GPT‑5.6 Sol, into the software that controls superconducting qubit experiments.
Beatriz Yankelevich, of MIT's Engineering Quantum Systems Group (EQuS), connected Codex to the lab software that coordinates measurements on an uncalibrated six-qubit chip — a standard part the group uses to benchmark its fabrication process. She supplied measurement-specific skills describing how to run and evaluate each experiment; using those and the chip's design targets, the agent selected measurement parameters, operated the hardware, analyzed the data, and either refined the measurement or passed the result to the next step.
When signals were clear, OpenAI says Codex completed a standard calibration sequence with little intervention, identifying qubit transition frequencies, calibrating control and readout pulses, and measuring coherence time. When signals were weak or noisy, it took longer to find suitable parameters and sometimes required guidance from an experienced researcher. OpenAI concedes experienced researchers may still find the best calibration settings faster than current models.
Characterizing one of these chips can take a researcher several days; the group now routinely delegates such measurements to agents. "I can have agents running measurements for many hours overnight or while I'm working in the cleanroom," Yankelevich said.
The post is a vendor account of a single lab's workflow. It reports no benchmarks, no error rates, no time-savings figures, and no comparison against other models or automation tooling.