--- url: "https://www.xcademia.com/news/gpt-5-6-sol-helps-automate-quantum-computing-experiments-with-codex" title: "GPT-5.6 Sol Helps Automate Quantum Computing Experiments With Codex" description: "OpenAI says GPT-5.6 Sol can automate routine quantum-chip measurements through Codex, helping researchers analyse results and design experiments." publishedAt: "2026-09-09T10:02:32.837+00:00" updatedAt: "2026-09-09T12:01:57.385051+00:00" type: news category: "ai-ml" source_name: "OpenAI - Applied AI" source_url: "https://openai.com/index/codex-quantum-computing-experiments/" tags: - "#OpenAI" - "#GPT56Sol" - "#Codex" - "#QuantumComputing" - "#QuantumAI" - "#SuperconductingQubits" - "#AIResearch" - "#AppliedAI" --- # GPT-5.6 Sol Helps Automate Quantum Computing Experiments With Codex > OpenAI says GPT-5.6 Sol, connected to laboratory software through Codex, can autonomously run routine superconducting-qubit measurements, helping researchers spend more time on experiment design and analysis. Source: **OpenAI - Applied AI** · 9 September 2026 **Q**uantum computing research can involve sophisticated hardware, but a large part of the work surrounding that hardware is software-driven. Researchers must repeatedly configure experiments, collect measurements, analyse signals, calibrate qubits and determine what measurement should happen next. OpenAI says this workflow provided an opportunity to test whether an AI agent could take over some of the routine experimental work. At MIT's Engineering Quantum Systems Group, graduate student Beatriz Yankelevich connected GPT-5.6 Sol to laboratory software through Codex. The system was used to run and refine measurements on superconducting quantum chips, allowing routine experiments to continue with less direct supervision. The result was not a replacement for the researcher. Instead, the experiment demonstrated how an AI agent can operate within a defined scientific workflow while leaving researchers more time for experiment design, analysis and planning. ## From AI model to laboratory agent Quantum processors use quantum bits, or qubits, rather than conventional bits. In the MIT group's work, the qubits are superconducting devices cooled to near absolute zero inside dilution refrigerators. Once a superconducting qubit chip has been fabricated, packaged and cooled, much of the interaction with the device happens through software. Microwave signals are used to control and probe the qubits, while returning signals are digitised and analysed. That software-controlled environment made the laboratory a practical testbed for an AI agent. Yankelevich provided Codex with measurement-specific skills that explained how individual experiments should be performed and evaluated. GPT-5.6 Sol could then select measurement parameters, operate the relevant laboratory software, analyse the resulting data and determine whether to refine the measurement or use its result in the next stage. ![info-1](https://0a515t3ure77wbvx.public.blob.vercel-storage.com/articles/1788948053263-info1--63-.webp) ## Why qubit calibration is suitable for AI agents Calibrating a quantum processor is not simply a matter of taking one measurement. The measurements are interdependent. One result can influence the parameters used in the next experiment. Researchers may need to determine a qubit's transition frequency, calibrate control and readout pulses, and establish how long quantum information is retained. The properties of a qubit can also drift, while unexpected physical behaviour can produce inconsistent signals. This creates a workflow that combines three characteristics that are useful for AI agents: - Repeated measurements - Software-controlled experimentation - Adaptive decision-making OpenAI says Yankelevich tested GPT-5.6 Sol on an uncalibrated six-qubit chip of a standard type used by the EQuS group for fabrication benchmarking. The agent used the chip's design targets and the supplied measurement skills to decide how to proceed. When experimental signals were clear, the system could complete a standard sequence with little researcher intervention. It identified qubit transition frequencies, calibrated pulses used for control and readout, and determined how long the qubit retained quantum information. ![info-2](https://0a515t3ure77wbvx.public.blob.vercel-storage.com/articles/1788948070038-info2--64-.webp) ## AI performs best when the workflow is clearly defined The experiment also exposed an important limitation. GPT-5.6 Sol had more difficulty when the experimental signals were weak or noisy. In those situations, it took longer to identify suitable measurement parameters and sometimes required guidance from an experienced researcher. That distinction is important. The case study does not show that an AI agent can independently solve every problem encountered during quantum experimentation. Instead, it demonstrates stronger performance in workflows where the objectives, procedures and evaluation criteria are sufficiently well defined. Ambiguous physical results remain more difficult. This creates a practical division of responsibilities. An AI agent can handle repetitive experimental operations and data-processing steps, while researchers remain responsible for interpreting uncertain results and deciding where the research should go next. ## Moving beyond routine measurements The work at EQuS extends beyond automated calibration. Yankelevich has built infrastructure that allows agents to participate in several parts of her research, including measurement, theory and chip design. For novel experiments, agents can be given narrower goals and can use their ability to write, modify and test code for experimental control, analysis and simulation. This creates a different model for scientific computing. Rather than using AI solely as an analysis assistant, researchers can connect an agent directly to parts of the experimental environment. The agent can modify code, test it against real measurements and continue through multiple stages of a defined task. The human researcher remains responsible for the higher-level scientific direction. ## Researchers can focus on higher-level scientific work According to OpenAI, the EQuS group now regularly uses agents for routine measurements on standard chips. Yankelevich said agents can run measurements for extended periods while she works on other research activities, with the ability to check results and intervene when necessary. The benefit is therefore not simply faster measurement. It is a change in how researchers spend their time. Instead of continuously monitoring every calibration step, researchers can devote more attention to interpreting results, designing experiments, planning subsequent work, reading research and writing. The case study also describes the possibility of multiple agents working on different problems simultaneously, while the researcher coordinates the broader research programme. ![info-3](https://0a515t3ure77wbvx.public.blob.vercel-storage.com/articles/1788948098215-info3--61-.webp) ## What this experiment says about AI and science The broader significance of the work is the connection between AI agents and physical experimentation. Many scientific workflows contain repetitive computational stages between a researcher and a physical instrument. If those stages can be expressed through software and governed by clear procedures, they can potentially become accessible to AI agents. The quantum computing experiment highlights that possibility. The agent was not simply asked to explain quantum mechanics or analyse a static dataset. It was connected to software that could interact with laboratory equipment, allowing it to participate in an iterative measurement process. For researchers, this could mean shifting AI from a passive analytical tool toward a more active laboratory assistant. However, the limitations are equally significant. The experiment showed that unclear or noisy measurements can still require experienced human judgement. The company did not provide specific information about broader performance across other quantum hardware platforms or experimental environments in this announcement. ## A practical model for AI-assisted laboratories The EQuS experiment points toward a model in which scientific AI systems operate within carefully defined boundaries. Routine tasks can be delegated to agents. Researchers can monitor progress and intervene when necessary. More complex scientific decisions remain under human supervision. This approach may be particularly relevant to experimental fields where instruments are software-controlled and where experiments consist of repeated measurement and adjustment cycles. The announcement highlights a broader industry shift toward AI systems that can interact with tools and execute multi-step workflows rather than simply generate text or answer questions. For enterprises and research institutions, the development could also demonstrate the value of connecting AI systems to existing software infrastructure instead of treating AI as a standalone interface. ## The bigger picture OpenAI's quantum computing case study provides an example of AI moving closer to the physical research environment. GPT-5.6 Sol, operating through Codex, was able to perform routine measurements on superconducting qubits, analyse the results and adapt subsequent measurements within a defined workflow. When signals became ambiguous, experienced researchers were still needed. That balance is likely to remain important as AI agents enter scientific laboratories. The immediate opportunity is not necessarily to remove researchers from the experimental loop. It is to reduce the amount of researcher time spent on repetitive operations and create more room for scientific judgement, creativity and experiment design. In quantum computing, where researchers can spend days characterising individual chips, even that shift could be meaningful. ## Original source https://openai.com/index/codex-quantum-computing-experiments/ ## Tags `#OpenAI` · `#GPT56Sol` · `#Codex` · `#QuantumComputing` · `#QuantumAI` · `#SuperconductingQubits` · `#AIResearch` · `#AppliedAI` --- ## About this content This Markdown news article is the citation-grade twin of [GPT-5.6 Sol Helps Automate Quantum Computing Experiments With Codex](https://www.xcademia.com/news/gpt-5-6-sol-helps-automate-quantum-computing-experiments-with-codex). It is published by **Xcademia** (UK Companies House 12322710) and is available for AI search engines and large language models to index, summarise, and cite. When citing or quoting, please attribute *Xcademia* and link back to the source URL above. - Source: https://www.xcademia.com/news/gpt-5-6-sol-helps-automate-quantum-computing-experiments-with-codex - Publisher: Xcademia — https://www.xcademia.com - Catalogue index: https://www.xcademia.com/llms-full.txt