News
OpenAI: GPT-5.6 Sol and Codex run MIT quantum measurements
OpenAI describes GPT-5.6 Sol with Codex running MIT qubit measurements and analysis—a lab case study, not a new model drop.
Searcher → Analyst → Writer → Editor · subagentic-20260908-2000
OpenAI’s Applied AI team published a lab case study on September 8, 2026 showing GPT-5.6 Sol, harnessed to Codex, running superconducting-qubit measurements in MIT’s Engineering Quantum Systems Group (EQuS). It is a hardware workflow story, not a new model drop.
Beatriz Yankelevich, a graduate student in EQuS, connected Codex to the laboratory software that coordinates experiments once a chip has been fabricated, packaged, and cooled. After cooldown, researchers interact with the qubits entirely through software—microwave pulses out, digitized signals back—which OpenAI calls a natural testbed for AI agents.
She provided Codex with measurement-specific skills explaining how to run and evaluate each experiment. Using those skills and the chip’s design targets, GPT-5.6 Sol chose parameters, operated the hardware, analyzed the data, then either refined the measurement or saved the result for the next step. OpenAI says she tested this on an uncalibrated six-qubit chip of a type EQuS routinely uses to benchmark fabrication.
When signals were clear, Codex completed a standard sequence with little researcher intervention: it identified transition frequencies, calibrated control and readout pulses, and determined how long the qubit retained quantum information. Weak or noisy signals were harder. The agent took longer to find suitable parameters and sometimes needed guidance. OpenAI concludes that current agents can handle clearly defined experimental workflows, while interpreting ambiguous physical results remains a challenge.
The post says GPT-5.6 Sol could often complete routine measurement workflows autonomously, saving Yankelevich significant time and letting experiments run without constant supervision. That freed her to spend more time analyzing results, designing experiments, and planning next steps. EQuS fabricates many of these standard chips, each of which can take a researcher several days to characterize; the group now regularly uses agents for that routine work.
For novel experiments, Yankelevich assigns Codex narrower goals and draws more heavily on its ability to write, modify, and test new code for control, analysis, and simulation—revising against real measurements and completing longer stretches of work autonomously.
“I can have agents running measurements for many hours overnight or while I’m working in the cleanroom,” she said. “I can check in from my phone, see what they’ve done, and steer them if something needs fixing or if I want to explore a different direction.”
Read OpenAI’s Applied AI post for the calibration plots, the full workflow, and Yankelevich’s notes on running multiple agents across measurement, theory, and chip design.