OpenAI's GPT-5 Accelerates Scientific Research Across Multiple Disciplines

New Research Shows How AI Could Transform Math, Physics, Cancer Research and More

A new paper shows ChatGPT-5 emerging as a tool that helps scientists test ideas, navigate literature and refine experiments A new report from OpenAI and a group of outside scientists shows how GPT-5, the company's latest AI large language model (LLM), can help with research from black holes to cancer‑fighting cells to math puzzles. Each chapter in the paper offers case studies: a mathematician or a physicist stuck in a quandary, a doctor trying to confirm a lab result. They all ask GPT-5 for help. Sometimes the LLM gets things wrong. Sometimes it finds a faster route to an already known result. But other times, with careful human guidance, it helps push the boundaries of what was previously known. In one experiment involving how waves behave around black holes, GPT-5 worked through the math to independently produce results that had previously been shown to be correct, showing it was capable of doing this level of scientific calculation. In another project involving nuclear fusion, GPT-5 developed a model that accelerated the research. "AI's ability to dramatically reduce the time required for coding -- compressing what would traditionally take days into mere minutes for the author -- has monumental implications for research practices," says Floor Broekgaarden, an astronomer at the University of California, San Diego, who was not involved in the study. If you're enjoying this article, consider supporting our award-winning journalism by subscribing. By purchasing a subscription you are helping to ensure the future of impactful stories about the discoveries and ideas shaping our world today. In another case, researchers studying immune cells used GPT-5 to interpret their data, and its explanation matched results the lab had already confirmed. "GPT-5 Pro can function as a true mechanistic co-investigator in biomedical research, compressing months of reasoning into minutes, uncovering non-obvious hypotheses, and directly shaping experimentally testable strategies," Derya Unutmaz, the doctor leading the project, wrote in the paper. The paper also announces several new math discoveries supported by GPT-5. Guided by human experts, it solved a long-standing problem posed in 1992 by mathematician Paul Erdős. It also produced a clearer rule showing the limitations of how computer systems make decisions; discovered another rule for how certain small patterns appear inside branching diagrams; and found a way to spot secret structures in a network as it grows. The discoveries are modest but appear to be genuine, and each was verified by human mathematicians. "I had not seen anything that impressive [in math] from an LLM before," says Ryan Foley, an astrophysicist at the University of California, Santa Cruz, who was not involved in the study. "I suspect LLMs are going to upend how theories are created, vetted and improved." He cautions, however, that AI tools still require significant prompting: "Humans are creative; AI is responsive. However, the rate of discovery should rapidly increase." Prithviraj Ammanabrolu, a computer scientist at the University of California, San Diego, who was not involved in the research, points out that the published work is more a series of case studies than a scientific paper because it doesn't provide enough details to repeat the experiments and doesn't offer counterfactual analysis involving different approaches. Despite these limitations, AI's ability to help with research "is still miles ahead of what was possible even a year ago, so the rate of progress is quite high," he says. "It shows future potential in enabling scientists to accurately mix together relevant prior results and draw new insights in novel ways." One of GPT-5's strengths is its ability to search vast quantities of scientific literature. For a math problem listed as unsolved online, it identified a solution in a paper from the 1980s. In another case, it found a few lines in a German paper from the 1960s that settled a problem. It easily navigated the language barrier and the differences in style between midcentury math writing and contemporary approaches. All of this might make GPT‑5 sound like a scientific genius, but the paper's authors are clear that it's not. Rather, in the right hands, it is a fast and tireless assistant that has read an impossible number of papers and never minds reworking a calculation. But human judgment is not optional, they stress. Researchers also caught it being confidently wrong, and it can misstate references, hallucinating nonexistent papers or failing to credit authors of real ones. "Human expertise remains crucial," Broekgaarden says. But AI "can take on myriad tasks -- collating data, summarizing research articles, and even performing complex calculations -- that previously demanded extensive time and effort from researchers." Numerous ways that AI will shape research remain to be seen. New AI models are released every few months. If general‑purpose chatbots that struggled with middle school math two years ago can now spot hidden structures in black-hole waves and suggest new approaches to cell therapy, who knows what their successors will achieve?

GPT-5 is speeding up scientific research, but still can't be trusted to work alone, OpenAI warns

GPT-5 supports researchers across disciplines, a study found. The model doesn't rival human researchers, however. The findings don't indicate AGI is coming soon. OpenAI's recently released model, GPT-5 is showing promise in advancing scientific discovery. While user reactions to the new model in ChatGPT were less than stellar, it appears to be making more headway as a research assistant. In a new paper published Thursday, OpenAI detailed the ways GPT-5 "accelerated" research across a variety of case studies -- albeit with some limitations. "Across these early studies, GPT-5 appears able to shorten parts of the research workflow when used by experts," the paper said. "It does not run projects or solve scientific problems autonomously, but it can expand the surface area of exploration and help researchers move faster toward correct results." Also: OpenAI tested GPT-5, Claude, and Gemini on real-world tasks - the results were surprising CEO Sam Altman and Chief Scientist Jakub Pachocki reiterated the company's science-forward goals during a livestream last month, in which they also discussed ambitious timelines for developing artificial general intelligence (AGI), which would theoretically be comparable to human ability. (Disclosure: Ziff Davis, ZDNET's parent company, filed an April 2025 lawsuit against OpenAI, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.) It's the first report from OpenAI for Science, a team of internal researchers and recently-hired external academics that the company announced in September. The paper was also supported by researchers from several labs and universities, including Vanderbilt, UC Berkeley, Columbia, Cambridge, Oxford, The Jackson Laboratory, and others. According to a blog accompanying the paper, OpenAI for Science aims to help researchers save time by using frontier models to test hypotheses and reveal insights from vast datasets. The results are early, but frontier models are evolving rapidly -- for now, researchers appear optimistic that AI will help us unlock novel, if incremental, discoveries. The paper highlighted several case studies in which GPT-5 helped with or advanced scientific endeavors in biology, math, and algorithmic decision-making. The model's contributions ranged from creating smaller-scale efficiencies -- like improving a proof for a mathematical theorem -- to larger breakthroughs. Also: AI models know when they're being tested - and change their behavior, research shows In one example of the latter, Jackson Laboratory scientists had spent months reading and experimenting in an immunology trial to eventually explain a change in immune cells. They gave GPT-5 unpublished data from the trial -- so as to ensure the model hadn't already been trained on it -- to see if it could come up with a similar conclusion. "GPT-5 identified the likely cause within minutes from an unpublished chart and suggested an experiment that proved it," OpenAI wrote. The implication is that medical researchers can involve frontier models earlier on in their experiments to improve treatments and understand diseases in minutes, not months. In another case study, GPT-5 helped a separate Jackson Laboratory team conduct a deep literature search that revealed connections between the team's newly-proven geometry theorem and other areas of math. GPT-5 efficiently flagged other areas the team could apply its findings to and surfaced reference material it hadn't encountered, including some in other languages. The model saved the researchers the task of manually reviewing literature for connections and broadened their knowledge base in the process. Also: Google's Antigravity puts coding productivity before AI hype - and the result is astonishing "These collaborations help us understand where the models are useful, where they fail, and how to integrate them into the scientific process -- from literature review and proof generation to modeling, simulation, and experimental design," the company wrote. Many of the paper's examples demonstrated that GPT-5 can rapidly reach existing scientific conclusions -- what OpenAI referred to in one case study as "independent rediscovery of known results." However, the paper also mentioned "four new results in mathematics (carefully verified by the human authors), underscoring that GPT-5 can solve problems that people have not yet solved." In one example, Columbia researcher Mehtaab Sawhney and OpenAI researcher Mark Sellke explored an existing number-theory problem from Hungarian mathematician Paul Erdős known as #848. It's marked "open," or unresolved, on a public site where users can contribute solutions -- not because humans haven't made headway solving it, but because those proposed solutions are scattered around in notes and textbooks, and not centralized or necessarily agreed upon. While GPT-5 didn't come up with an entire answer for #848 out of thin air, which really would have rivaled human ability, it was able to identify the final proof's missing step. "Human comments on the site had already outlined much of the structure; GPT-5 proposed a key density estimate, and Sawhney and Sellke corrected and tightened it into a complete proof that closed the problem," OpenAI wrote. In another study, GPT-5 came up with two proofs -- one previously proven, one new -- for a graph theory problem, "relying on a different and more elegant argument than the original human proof." As with other examples, the researchers were able to verify and adopt GPT-5's suggestion. Given how quickly frontier models have evolved in the last three years, the researchers believe "these contributions are modest in scope but profound in implication." Despite these strides, GPT-5 wasn't foolproof. OpenAI recommended it only be used with continued oversight from experts. "GPT-5 can sometimes hallucinate citations, mechanisms, or proofs that appear plausible; it can be sensitive to scaffolding and warm-up problems; it sometimes misses domain-specific subtleties; and it can follow unproductive lines of reasoning if not corrected," OpenAI noted. For those reasons and others, the paper doesn't suggest AI tools replace current scientific research methods just yet. Advocating for a partnered approach, OpenAI said that while the core tools of science, including simulators and algebra systems, are crucial to maintaining precision and efficiency, the reasoning abilities advanced models provide are a valuable step forward. "Where specialized tools exist, we want to use them; where general reasoning is required, we build models designed to handle it," the company wrote. "Both paths reinforce each other." The paper emphasized that scientists should remain in charge by defining questions, critiquing concepts, and checking results -- GPT-5, in this case, provides speed and reach to scale that expertise. Like basic forms of prompt engineering, OpenAI noted that scientists must learn to communicate with GPT-5 for the best results, and that ultimately, "productive work often looks like dialogue" between humans and the model -- a common theme across many AI tools and assistants pitched as copilots or drafting companions, though those are often built for simpler consumer tasks. Also: 10 ChatGPT prompt tricks I use - to get the best results, faster The paper suggested that GPT-5 is at most approaching the level of a research partner, with some limitations. In another use case, combinatorialist Tim Gowers gave the model several tough questions he was working on and asked it for feedback, critique, and counterexamples. GPT-5 found flaws and offered simpler arguments in some instances, but stalled out or didn't make any progress in others. "Gowers' overall conclusion was that the model is already useful as a very fast, very knowledgeable critic that can stress-test ideas and save time, even though it does not yet meet his bar for full co-authorship," OpenAI concluded. Ultimately, the OpenAI for Science paper exemplifies GPT-5's strengths in refining and assisting -- filling in gaps rather than going toe-to-toe with human minds. While OpenAI acknowledged that models have surpassed just summarizing existing information, that doesn't mean the company is prepared to say GPT-5 is an indicator of AGI. "We don't view these results as signs that we are close to AGI or a fully capable 'research intern,'" the company told ZDNET in a statement, referring to Altman's comment in last month's live stream that OpenAI will release a model with intern-equivalent research capabilities by September 2026. "Benchmarks across the field are saturating, so we are putting more of an emphasis on testing a model's capabilities, including how the models work in scientific workflows. That gives us a clearer picture of actual capability and limitations."

OpenAI says new GPT-5 model speeds up research in maths and science

OpenAI said its latest artificial intelligence model GPT-5 is accelerating research in mathematics, biology and physics, as AI groups race to produce tools for scientists in the hunt for new revenue streams. The $500bn start-up on Thursday published a paper that showed its large language model helped a mathematician at Columbia University crack an unsolved and notoriously difficult maths equation called the Erdős number theory problem. OpenAI said GPT-5 also identified a change in human immune cells within minutes, which scientists had spent months trying to solve. The model suggested an experiment that researchers were able to test and confirm to be correct. "If we can build these tools and put them in the hands of scientists all over the world, that can help them do the next 25 years of scientific research in five years instead," said Kevin Weil, OpenAI's vice-president of science. The development comes as tech groups push into the science industry, as they bet on the potential for AI to speed up drug discovery and uncover new materials. Last month, Anthropic revealed plans to integrate its Claude chatbot into tools used by researchers and life sciences companies. Google has also unveiled a "co-scientist" tool that could help researchers come up with new hypotheses, and last month said its open Gemma model had helped discover a new potential cancer therapy pathway. OpenAI set up a new science unit in October and has hired Alex Lupsasca, a theoretical physicist known for his work on black holes, as a research scientist. It plans to build an "automated AI research intern" by next September and a fully automated AI research tool by March 2028. The group has struck a series of massive deals this year with companies including Nvidia, Oracle, AMD and Broadcom as part of a push to secure the huge computing power it needs to develop the technology's potential. "AI really is now capable of being used to advance science for society," said Ruairidh Battleday, an AI researcher at Stanford University who has studied how the cutting-edge technology can drive scientific discoveries. But he noted that current models were not yet comparable to a "fully autonomous AI scientist", but more a co-pilot that, when "guided by a skilled scientist, has access to a really impressive range of the literature and set of quantitative tools". OpenAI said GPT-5 was good at doing in-depth literature searches to connect dots across languages and fields of research. It also independently rediscovered findings from new research not included in its training data. However, the paper highlighted that GPT-5 still hallucinated or made things up, and human experts needed to define problems and correct assumptions and outputs. "It can help you get more done, move faster, try more things, maybe uncover things that you might not have on your own. But you have to check. It doesn't absolve you from rigour," said OpenAI's Weil. Verifiable problems such as coding, maths and formal logic were "extremely well suited for LLMs," said Jakob Foerster, an associate professor at the University of Oxford, who has also built AI tools for scientific research. "Unfortunately, much of the progress seen here is unlikely to generalise to rather mundane real-world tasks in business applications," he added.