GPT-5 lowers the cost of cell-free protein synthesis

GPT-5 lowers the cost of cell-free protein synthesis | OpenAI

February 5, 2026

GPT‑5 lowers the cost of cell-free protein synthesis

Working with Ginkgo Bioworks, we created an AI-driven autonomous lab and achieved a 40% reduction in protein production cost.

Loading…

Share

We’ve seen rapid progress from AI in fields like math and physics, where ideas can often be evaluated without touching the physical world. Biology is different. Progress runs through the lab, where scientists run experiments that take time and money.

That’s starting to change. Frontier models can now connect directly to lab automation, propose experiments, run them at scale, learn from the results, and decide what to do next. In much of life science, the bottleneck is iteration, and autonomous labs are built to remove that constraint.

In earlier work, we showed that GPT‑5 could improve wet-lab protocols through closed-loop experimentation. Here, we show that the same approach can reduce the cost of protein production.

We partnered with Ginkgo Bioworks⁠ to connect GPT‑5 to a cloud laboratory—an automated wet lab run remotely through software, where robots execute experiments and return data—and used that lab-in-the-loop setup to optimize a widely used biological process: cell-free protein synthesis (CFPS). Over six rounds of closed-loop experimentation, the system tested more than 36,000 unique CFPS reaction compositions across 580 automated plates. After being provided access to a computer, a web browser, and access to relevant papers, GPT‑5 took three rounds of experimentation to establish a new state of the art in low-cost CFPS, achieving a 40% reduction in protein production cost (and a 57% improvement in the cost of reagents), including novel reaction compositions that are more robust to reaction conditions common in autonomous labs.

Why cell-free protein synthesis matters

Cell-free protein synthesis (CFPS) is a way to make proteins without growing living cells. Instead of putting DNA into cells and waiting for them to produce a protein, CFPS runs the protein-making machinery in a controlled mixture. That makes it a practical tool for rapid prototyping and testing as scientists can run many experiments quickly and measure results the same day.

Proteins are a big part of what modern biology delivers. Many important medicines are based on proteins. Many diagnostics and research assays depend on proteins. In industrial settings, proteins act as enzymes that make chemical processes cleaner and more efficient. Proteins are even found in your laundry detergent. When protein production becomes faster and cheaper, scientists can usually test more ideas sooner, and reduce the cost of turning early research into something that people can benefit from everyday.

CFPS is already useful for that kind of iteration. The bottleneck is that it is tricky to optimize and gets expensive at scale.

Cell-free protein synthesis is difficult to optimize and costly

Cell-free protein synthesis requires complex, interacting ingredients: the DNA template encoding the protein to be made, the cell lysate (the soup of cellular machinery from inside cells), and a large number of biochemical components ranging from energy sources to salts. It is incredibly difficult to reason about the system as a whole, and many⁠ previous⁠ studies⁠ have applied different types of machine learning to reduce protein production cost.

Standard cell-free protein synthesis (CFPS) formulations and commercial kits are often priced for human-paced work. Autonomous labs can run thousands of reactions in the time a human team might run dozens. At that scale, the cost of reagents becomes the limiting factor.

CFPS is also difficult to optimize by intuition alone. It’s a mixture of many interacting components. Small changes can matter, but the direction of the effect isn’t always obvious, and the best combinations can be hard to find without running a lot of experiments. Prior approaches have reduced costs, but progress tends to be incremental because exploring the space thoroughly is labor-intensive.

Connecting GPT‑5 to a robotic lab

We paired GPT‑5 with Ginkgo Bioworks’ cloud laboratory to form a closed-loop autonomous system for cell-free protein synthesis (CFPS) optimization.

GPT‑5 designed batches of experiments. The lab executed them. The results were fed back to the model. The model used that data to propose the next round. We repeated that cycle six times.

GPT‑5 designed batches of experiments in a standard 384-well plate format, and ran them on Ginkgo Bioworks’ cloud laboratory. Once the experiments finished, the cloud laboratory pushed the data back to GPT‑5, where the model analyzed the outcomes, generated new hypotheses, and designed the next round of experiments.

To keep the loop grounded in what an autonomous lab can do, we added strict programmatic validation before any experiment ran. That validation enforced that AI-designed experiments were physically executable on the automation platform. It prevented “paper experiments” that look plausible in text but can’t be carried out in a robotic workflow.

Across the full run, the system executed more than 36,000 CFPS reactions across 580 automated plates. This scale matters because it’s what lets patterns emerge. In biology, single experiments are noisy. Throughput and iteration are how you separate signal from random noise. Once GPT‑5 had access to the relevant paper and tools, it took three rounds of experimentation and two months to establish a new state of the art: 40% lower protein production cost compared to the best prior baseline⁠.

Ginkgo Bioworks’ reconfigurable automation carts. Credit: Ginkgo Bioworks

What we learned

We found that the improvements came from identifying combinations that work well together and that hold up in the realities of high-throughput automation.

We found that GPT‑5 identified low-cost reaction compositions that humans had not previously tested in this configuration. Cell-free protein synthesis (CFPS) has been studied for years, but the space of possible mixtures is still large. When you can propose and execute thousands of combinations quickly, you can find workable regions that are easy to miss with a manual workflow.

We also found that high-throughput, plate-based experiments often differ from manual, bench-top experiments. Oxygenation can be lower in high-throughput reaction formats. Mixing and geometry can be different. Most CFPS…