Applications open November 23 as new research finds scientists report the largest time savings in coding, while checking AI outputs remains a leading difficulty
OpenAI’s next academic research grant round will provide ChatGPT Plus access for recipients and up to two collaborators each.
OpenAI will offer 50,000 new grants through ChatGPT for Academic Researchers, expanding free access as a study of its summer pilot finds scientists estimate that equivalent research work would have taken about 1.9 times as long without AI.
Applications open Monday, November 23. Each grant will provide ChatGPT Plus access for the recipient and up to two collaborators, potentially supporting 150,000 researchers in this round. OpenAI has doubled its overall ambition for the program to reach 200,000 researchers worldwide.
The expansion changes both selection and access. Grants will go to the first eligible applicants rather than being allocated by lottery, and new recipients will receive Plus instead of the Pro subscription offered during the pilot.
Alongside the announcement, OpenAI has published Science at the Frontier: How Scientists Use ChatGPT. Its findings combine observed usage with a survey, showing where scientists request assistance and where they believe it saves time. They do not establish that AI causes faster scientific discovery.
First-come grants replace the pilot lottery
Since the program began in July, 20,000 researchers in 33 countries have joined, according to OpenAI. Participants receive access to its models and training on using ChatGPT in research.
The company says its decision to offer Plus for new grants follows a review of pilot usage, which found that most participants would have remained within Plus usage limits. The change is intended to extend access to more research teams.
Eligible researchers already on the waitlist will receive an email before applications open with instructions for activating their accounts. Data shared through the program is not used to train OpenAI’s models by default.
Debbie Bard, Lead for Academic Research Partnerships at OpenAI, announced the expansion on LinkedIn and invited researchers to share what they were finding useful and where they were encountering difficulties. OpenAI says that feedback will inform future rounds and the support it provides.
Coding produces the largest reported time savings
The research paper, by Chris Liao, Carl Shan and Aaron Chatterji, examines 580,186 scientific messages from 1,819 scientists using ChatGPT and Codex between August 20 and September 22, 2026.
A separate survey ran from September 30 to October 7. Of 270 valid completed responses, 251 participants consented to inclusion in the analysis.
Across eight research activities, respondents estimated that work completed with AI available would have taken about 1.9 times as long without it to reach the same quality. The largest reported gain was in research software and code development, where the equivalent work was estimated to require 2.83 times as long without AI. Administrative work had the smallest reported gain, at 1.35 times.
These are participants’ estimates of time required, rather than results from a controlled comparison. Each research activity counted equally in the overall average, and scientists reporting no AI use for an activity were treated as having no productivity gain for that activity.
The paper also describes interview accounts in which scientists used coding assistance to build specialized research tools or enable data collection that would otherwise have been impractical. Those accounts suggest possible changes in the work researchers can attempt, alongside savings on existing tasks.
Learning and writing dominate scientific use
Despite attention to AI’s ability to conduct research, most activity in the study involved helping scientists understand, communicate and execute their own work.
Reading and learning accounted for 28% of the average scientist’s scientific messages, followed by writing and communicating at 27%. Analyzing evidence and theories represented 17%, and research software development accounted for 14%. Together, those four activities made up about 85% of scientific messages.
The figures describe the distribution of messages, not working hours. The researchers calculated each scientist’s share of messages in an activity and then averaged those shares, giving scientists equal weight.
At the level of individual tasks, explaining scientific concepts was the most common use. Almost 90% of scientists sent at least three messages requesting that kind of assistance during the study. About 82% used the tools at least three times to draft manuscripts.
OpenAI’s gives examples of research applications. Shuqi Li, a researcher at Brigham and Women’s Hospital and Harvard Medical School, is exploring whether ChatGPT can connect laboratory and patient data to support antibiotic development and the search for drug targets that could shorten tuberculosis treatment.
Arthur Porto, Assistant Curator of AI Biodiversity at the Florida Museum of Natural History, and his team use AI tools to turn photographs of museum specimens into searchable databases. They also created a 3D butterfly model that reproduces the specimen’s changing wing colors as it is rotated.
Scientists shape ideas and check the results
Developing ideas accounted for only 5% of scientific messages, but 68% of scientists met the study’s threshold for ideation use, defined as sending at least three messages about developing research ideas.
The interactions took several forms. In 35% of ideation messages, scientists brought a specific question for ChatGPT to work with. In 36%, they supplied an idea and asked the system to refine it. Requests for ChatGPT to suggest a research question accounted for 23%, while 6% could not be categorized.
Those percentages describe roles within ideation exchanges, averaged across scientists. They do not represent separate groups of researchers.
Continued involvement also appeared after scientists delegated tasks. Across eligible initial requests, scientists followed up to clarify, check or revise AI output about 32% of the time. Follow-up rates were highest in writing and communicating, at 38%, followed by research software development and analysis, both at 34%.
The measure captured whether the next scientific message in the same conversation, within seven days, sought clarification, checking or revision. It does not establish whether the original response was accurate or whether subsequent checks resolved a problem.
The paper identifies the time needed to check or correct outputs as the most commonly reported difficulty in obtaining value from AI.
Requests for end-to-end autonomous research remained uncommon. Around 15% of scientists made at least one request asking AI to coordinate multiple stages of research, but these requests accounted for just 0.4% of the average scientist’s scientific messages. The authors also found limited sustained use of that approach.
Optimism about discovery meets staffing and funding constraints
Among the 250 scientists who answered the discovery-timeline question, 98% expected AI to accelerate discovery in their field.
Respondents considered a hypothetical discovery that would take ten years without AI. About 22% expected it could happen within three years with AI, while 50% expected it within four to seven years.
Asked to explain that optimism, scientists most often described faster research execution. Other responses concerned finding more promising ideas and pursuing a broader range of questions. These were expectations about future progress, rather than observations of completed discoveries.
Existing constraints remained substantial. Limited time or staffing was selected by 51.8% of survey respondents, while 43.8% identified funding. Overall, 71% selected at least one of those two constraints; respondents could choose up to three.
The authors caution that participants were a selected group who had applied for expanded AI access and, for the survey, chosen to answer questions about their experience. The findings may therefore not generalize to scientists more broadly.
For privacy, the research team analyzed categories produced by automated classifiers rather than reading the underlying messages or ChatGPT responses. Usage data records requests to AI, not the quality or eventual scientific value of the resulting work.
Other access routes serve universities and specialist labs
The expansion sits alongside separate institutional and specialist offerings. ChatGPT Edu allows universities to provide access for research, teaching and learning, with data analysis capabilities and administrative controls.
For biology researchers, GPT-Rosalind combines specialized biological reasoning with scientific tools and databases. Its separate Rosalind Academic Plan offers eligible laboratories at U.S. universities up to 30 seats at no cost for up to 12 months, subject to approval through OpenAI’s trusted-access process.
OpenAI has also contributed $5 million to a Simons Foundation initiative supporting mathematicians in examining and building on AI-generated results.
The next ChatGPT for Academic Researchers application round opens November 23, with 50,000 grants available to eligible applicants on a first-come, first-served basis.
ETIH Innovation Awards 2027 open for entries
The ETIH Innovation Awards 2027 are now open for entries, with 26 categories recognising companies, institutions, projects and individuals making a measurable impact across education technology. The global awards span AI in education, digital learning, higher education, skills and workforce, inclusion, education operations, evidence and impact, with entries managed through ETIH’s new online awards platform.
Entries are open worldwide, with the early-bird deadline on Friday 18 December 2026 and the final entry deadline on Friday 26 February 2027 at 23:59 UK time.
Explore the 2027 categories and judging criteria, read the How to Enter guide or start an entry through the ETIH Innovation Awards 2027 platform.