Autonomous calibration and measurement of superconducting qubits were demonstrated by linking laboratory software to GPT-5.6 Sol via Codex, allowing the AI to execute and optimize experiment workflows with limited human input.

OpenAI and MIT claim GPT-5.6 Sol with Codex conducted end-to-end quantum experiments on a six-qubit chip, handling parameter inference, hardware control, data processing, and adaptive decision-making.

Caveats note that AI performance on weak signals can be clumsy and human oversight remains necessary for interpreting anomalous results, reflecting typical caution in secondary reporting.

Calibration remains extremely challenging due to decoherence sensitivity, parameter drift, and the need for precise nanosecond-scale microwave control at millikelvin temperatures in dilution refrigerators.

The achievement is framed as a major shift toward automated scientific research, with roles shifting toward high-level design and real-time closed-loop experimentation, while stopping short of claiming full replacement of human researchers.

Researchers supplied the AI with experimental context, chip designs, measurement skills, and specialized skill packs to guide measurements and evaluation criteria.

The workflow combines theory, simulation design, hardware control, data fitting, and iterative optimization, dramatically shortening traditional weeks-long research cycles to hours or minutes.

The report includes broader AI breakthrough claims and references to high-profile mathematical problems, while noting external skepticism and competing claims about those assertions.

AI achieved autonomous operation across the experiment, data analysis, qubit calibration, and frequency determination with minimal human intervention—humans stepped in only four times across 40 measurements.

Remote monitoring enables progress checks from a phone or in the cleanroom, freeing researchers for data analysis, experiment design, and planning; Codex coordinates multiple agents to handle measurement, theory, and design tasks for long runs.

The system enables autonomous interdependent measurements on a six-qubit chip, adapting later steps based on prior results to manage drift and noise.

GPT-5.6 Sol identifies transition frequencies, calibrates control and readout pulses, and determines information retention times using microwave signals and digitized readouts from the dilution fridge.