OpenAI: GPT-5.6 Sol helps run quantum computing experiments

OpenAI published how MIT Engineering Quantum Systems graduate student Beatriz Yankelevich connected GPT-5.6 Sol via Codex to lab software that runs and refines routine measurements on superconducting qubit chips.

A named coding agent is inside real quantum-chip tuning at MIT. Clear, routine measurements often ran with little help. Weak or noisy signals still needed an experienced researcher.

OpenAI published “How GPT-5.6 Sol helps run quantum computing experiments.” That company page is the filing event.

The subject on the page is Beatriz Yankelevich, a graduate student in MIT’s Engineering Quantum Systems Group (EQuS), using GPT-5.6 Sol harnessed to Codex on superconducting-qubit workflows. A qubit is the basic unit of quantum information. OpenAI says the chips are cooled near absolute zero in dilution refrigerators and controlled with microwave software; once fabricated, packaged, and cooled, researchers interact with the chip entirely through software.

OpenAI says connecting Codex to the lab software that coordinates experiments let the agent run measurements, analyze results, and decide what to try next. Yankelevich, as published by OpenAI, found GPT-5.6 Sol could often complete routine measurement workflows autonomously, freeing time for analysis and experiment design. Treat the autonomy and time-saved claims as company and lab framing — not an independent bench audit.

The testbed on the page is an uncalibrated six-qubit chip of a standard type EQuS uses to benchmark fabrication. Six-qubit means six quantum bits on one chip. Yankelevich gave Codex measurement-specific skills for how to run and evaluate each experiment. Using those skills and the chip’s design targets, Sol chose parameters, operated the hardware, analyzed the data, then refined the measurement or saved the result for the next step.

When signals were clear, OpenAI says Codex completed a standard calibration sequence with little intervention — identifying transition frequencies, calibrating control and readout pulses, and determining how long each qubit retained quantum information.

When signals were weak or noisy, Sol took longer to find workable parameters and sometimes needed guidance from an experienced researcher. OpenAI’s framing: current agents can handle clearly defined experimental workflows; interpreting ambiguous physical results remains hard. The company also notes experienced researchers may still find the best calibration settings faster than current models.

EQuS now regularly uses agents for routine measurements, OpenAI says. Yankelevich, quoted on the page: she can leave agents running measurements for many hours overnight or while she works in the cleanroom, check in from her phone, and step in if something needs fixing or she wants a different direction. That quote is hers as published by OpenAI.

This is a workflow case study on GPT-5.6 Sol plus Codex inside an MIT lab. It is not a new OpenAI quantum product SKU, a commercial lab robot, a Millennium-prize-style breakthrough, or a claim that Sol invents new physics. This filing does not invent chip yields, error rates, or a newly discovered qubit property beyond the page.