How GPT-5.6 Sol Helps Operate Quantum Computing Experiments

OpenAI collaborated with a research team at MIT to connect GPT-5.6 Sol and Codex to laboratory software, successfully automating the routine measurement and calibration processes of superconducting qubits. In quantum chip experiments that previously required hundreds to thousands of preliminary measurements and months of preparation, AI agents performed repetitive tasks, significantly reducing researchers' time.

Autonomous Calibration Workflow

The research team validated GPT-5.6 Sol's capabilities on an uncalibrated 6-qubit chip used for manufacturing process benchmarking. When provided with measurement-specific skills in Codex, the model selected measurement parameters based on chip design targets, operated the hardware, and analyzed data. In cases with clear signals, it completed standard measurement sequences such as resonance frequency identification, pulse calibration, and quantum information preservation time determination with minimal intervention.

Achievements and Limitations

The agent can independently perform measurements for hours, even overnight or while researchers are in the cleanroom, allowing progress to be monitored and directed via smartphone. This enables researchers to focus on high-value tasks such as interpreting results, designing experiments, and high-level planning.

However, limitations exist. Performance degrades in environments with weak signals or high noise, increasing the time required for appropriate parameter exploration, and interpreting ambiguous physical results still requires guidance from experienced researchers. Current AI models have been found to be slower than human experts in identifying optimal calibration settings.