GPT-5.6 Sol Autonomously Runs Quantum Experiments

An MIT researcher has demonstrated that frontier AI models can do far more than answer questions. By pairing GPT-5.6 Sol with Codex, the researcher was able to autonomously run quantum computing experiments, analyze the resulting data, and even calibrate qubits without constant human supervision. The setup handles multi-step experimental workflows that previously demanded extensive manual intervention, marking a notable shift in how scientific work can be conducted. The system works by combining reasoning with tool use. Instead of simply suggesting an experiment, the model can plan a sequence of actions, execute them through connected tools, interpret complex data, and then iterate based on what it learns. That feedback loop is what makes the approach feel less like a chatbot and more like a genuine scientific collaborator. Quantum computing is a particularly demanding test case. Qubit calibration is delicate, data-heavy, and repetitive, making it a natural fit for automation but a hard problem for AI to master. The fact that GPT-5.6 Sol and Codex could manage it suggests that similar agentic workflows could be applied across physics, chemistry, and biology. The work fits into a broader push to apply AI agents to scientific discovery. Researchers increasingly see large language models not as replacements for scientists but as tireless assistants capable of handling the tedious, iterative parts of experimentation. If the approach scales, it could accelerate research cycles and free human experts to focus on higher-level questions. For now, the MIT case study stands as a compelling proof of concept that autonomous AI-driven experimentation is no longer purely theoretical.