GPT-5.6 Sol Automates 6-Qubit MIT Experiments, Cutting Days of Calibration Work
Updated
Updated · OpenAI · Sep 8
GPT-5.6 Sol Automates 6-Qubit MIT Experiments, Cutting Days of Calibration Work
1 articles · Updated · OpenAI · Sep 8
Summary
MIT graduate researcher Beatriz Yankelevich found GPT-5.6 Sol, linked to Codex and lab software, could autonomously run routine measurements on an uncalibrated six-qubit superconducting chip.
Using predefined measurement skills and chip design targets, the agent chose parameters, operated hardware, analyzed data, and either refined the test or passed results into the next calibration step.
Clear signals let Codex complete standard workflows with little human intervention, identifying transition frequencies, calibrating control and readout pulses, and measuring qubit coherence times.
Weak or noisy signals still slowed the system and sometimes required expert guidance, showing current agents handle well-defined workflows better than ambiguous physical results.
EQuS now regularly uses agents on standard chips that can take researchers several days to characterize, freeing staff to supervise overnight runs and focus on analysis, experiment design, and code development.