OpenAI has just announced the activities of its newly established internal team, OpenAI for Science. The organization has launched a complimentary LLM-driven tool for researchers named Prism, which integrates ChatGPT into a text editor for composing scientific articles.
The aim is to position ChatGPT prominently within the software that researchers employ to document their findings, similar to how chatbots are now integrated into mainstream coding environments. It’s vibe coding, but tailored for scientific use.
Kevin Weil, leading OpenAI for Science, himself draws this comparison. “I believe that 2026 will serve for AI and science what 2025 did for AI in software engineering,” he remarked at a press event yesterday. “We are beginning to witness that same kind of transitional shift.”
OpenAI asserts that approximately 1.3 million researchers worldwide submit over 8 million inquiries weekly to ChatGPT concerning advanced scientific and mathematical topics. “This indicates that AI is transitioning from mere curiosity to being a fundamental aspect of researchers’ workflows,” Weil stated.
Prism is a response to this user behavior. It can also be interpreted as an effort to retain more researchers within OpenAI’s ecosystem in a competitive market filled with alternative chatbots.
“I primarily utilize GPT-5 for coding,” shares Roland Dunbrack, a biology professor at the Fox Chase Cancer Center in Philadelphia, who has no affiliation with OpenAI. “Occasionally, I pose scientific questions to LLMs, mainly hoping they can retrieve information from the literature more swiftly than I can. It previously generated fictitious references but it seems to have reduced that significantly.”
Nikita Zhivotovskiy, a statistician at the University of California, Berkeley, reports that GPT-5 has already become a vital resource in his work. “It occasionally aids in refining my paper’s text, correcting mathematical errors or issues, and provides generally valuable feedback,” he notes. “It proves extremely beneficial for swiftly summarizing research papers, enhancing engagement with scientific literature.”
By merging a chatbot with a routine software tool, Prism follows a pattern established by products like OpenAI’s Atlas, which integrates ChatGPT into web browsers, as well as LLM-enhanced office applications from companies like Microsoft and Google DeepMind.
Prism incorporates GPT-5.2, the company’s top model for tackling mathematical and scientific challenges, within an editor used for composing documents in LaTeX, a widely recognized coding language that researchers utilize for formatting papers.
A ChatGPT interaction box is positioned at the bottom of the screen, beneath the view of the article in progress. Researchers can request assistance from ChatGPT for various tasks. It can aid in drafting text, summarizing relevant articles, organizing citations, converting images of whiteboard notes into equations or diagrams, or discussing hypotheses or mathematical proofs.
It’s evident that Prism could be a significant time saver. However, it’s also apparent that many may be left wanting, especially following weeks of high-profile social media discussions from the organization’s researchers about how proficient GPT-5 is at resolving math problems. Science is inundated with AI oversights: Won’t this exacerbate the situation? Where is OpenAI’s fully automated AI researcher? And when will GPT-5 lead to a remarkable discovery?
That’s not the goal, according to Weil. He would love to see GPT-5 make a groundbreaking discovery. However, he doesn’t believe that will have the most substantial effect on science, at least in the short term.
“I think more significantly—and with absolute certainty—there will be 10,000 advancements in science that perhaps wouldn’t have occurred or wouldn’t have occurred as rapidly, and AI will have played a role in that,” Weil told MIT Technology Review in an exclusive interview this week. “It won’t be a dazzling highlight—it will merely represent a gradual, cumulative acceleration.”
