GPT-5.6 Sol runs quantum computing experiments at MIT

Quantum computing experiments typically involve months of preparation and thousands of preliminary measurements. Yankelevich’s work shows that AI agents can now handle well-defined calibration routines, which is a first step toward more autonomous research workflows.

The MIT group uses standard chips to benchmark their fabrication process, so characterization is repetitive. By connecting Codex to lab software, agents can run these measurements overnight or while researchers work elsewhere, as Yankelevich does from her phone.

The test case is whether AI can also interpret ambiguous physical results. Current models still struggle with noisy signals, so experienced researchers remain essential for novel or complex setups. If agents improve in handling such uncertainty, they could take on even larger portions of experimental work.