100,000 university researchers get free access to the flagship model for one year!
On July 29, OpenAI delivered a major gift to the academic community, officially launching "ChatGPT for Academic Researchers" to provide free access to cutting-edge models for 100,000 researchers.
Institutions such as École Normale Supérieure (ENS), which has produced numerous Nobel and Fields Medal laureates, and the Institute for Advanced Study (IAS) where Einstein once worked, were among the first selected institutions.
Eligible researchers can access GPT-5.6 Sol Pro, OpenAI’s most powerful flagship model to date.
Let me clarify: the target of 100,000 is for 2027; this summer, we will initially release 10,000 spots.
An entire research production line was packaged and sent in.
What can this system change for a researcher?
In the morning, you can use ChatGPT to read literature, question an idea, and casually generate several hypotheses.
In the afternoon, you can use Codex to write code, run data, and perform formal verification.
When it comes to applying for funding, writing grant proposals, or drafting manuscripts, let ChatGPT Work handle it.
This isn't just "adding another chat window"—it's consolidating your entire research workflow, which previously scattered across editors, literature software, terminals, and chat windows, into a single unified workspace.
The approved researchers have received exactly this full suite: ChatGPT, ChatGPT Work, Codex, along with an expanded Deep Research feature, higher usage limits, and a larger context window.
Combine over 75 life sciences skills: from genetics, genomics, and single-cell analysis to protein modeling and drug discovery.
In addition, there are numerous connectors available—literature libraries, public genomic and clinical databases, satellite imagery—all ready to use once integrated with Zotero and GitHub.
From Zotero and GitHub to Boltz and NGS analysis, commonly used research tools are bundled into a single workspace.
The models are specialized: GPT-5.6 Terra handles daily tasks, Luna runs lightweight tasks, and Sol tackles the toughest scientific and mathematical challenges.
On FrontierMath Tier 4, which measures research-level mathematical reasoning, Sol achieved 83%, compared to 72.5% for the previous generation, GPT-5.5; on GeneBench Pro, which tests complex biological data analysis, Sol Pro solved 31.5% of the problems.
From literature review and coding to running analyses and final manuscript submission, every step of the researcher’s process is integrated into a single system.
OpenAI's goal is clear: move your entire research process into ChatGPT.
1.3 million people per week are using ChatGPT for research.
Scientists have actually been using it this way for a long time.
According to official OpenAI data, approximately 1.3 million people per week use ChatGPT for advanced science and mathematics, generating around 8.4 million messages.
More intuitive signals in the math community: the number of math papers on arXiv acknowledging ChatGPT rose from just 14 in February to 100 in the first three weeks of July.
Number of arXiv mathematics papers acknowledging ChatGPT: increased from 14 in February to 97 in June; by only the first three weeks of July (through the 21st), it had already reached 100. (Source: OpenAI)
Moreover, the more aggressively you use it, the more confident you are in entrusting it with major tasks:
Among scientists in the top 20% for AI usage, nearly 7% dare to delegate tasks estimated to take more than four hours entirely to AI; among their peers, the proportion is only 3.5%, exactly half.
This shows that once a tool becomes easy to use, people naturally delegate increasingly heavier tasks to it. Habits are formed gradually in this way.
Seeing this data, it’s not hard to understand why OpenAI is giving away something so expensive for free to 100,000 scientists.
This workspace does not default to using your content to train models and provides robust enterprise-grade privacy protection—it’s not after your data, but your habits.
When a postdoc diligently completes a full year of research within this workflow, all their skills in searching, analyzing, and writing become fully integrated with ChatGPT and Codex.
By the time the free migration window closes a year later, the cost of moving will be so high that no one will bother.
Some industry observers have bluntly stated: the lab director should truly pay attention to which tool their team is quietly making the default option after booting up a computer.
This choice will continue to compound indefinitely after the free period ends.
Three limitations of a free lunch
This is free, but there are three limitations.
First, it’s not unlimited. The official statement clearly indicates that the usage limit is similar to ChatGPT Pro; exceeding it requires purchasing additional credits at your own expense.
Second, it does not include an API. The self-service plan explicitly states: no OpenAI API credits included.
Third, and most importantly: it is not open source, and no model weights have been provided.
Researchers are granted a 12-month product license, allowing them to access the model but not view its weights or modify it.
Moreover, the门槛 to enjoy this free lunch is also quite high.
Must be a research faculty member or postdoctoral researcher at a university, verified through SheerID, residing in a supported country, and must submit a paper published within the last three years on arXiv, bioRxiv, or ChemRxiv with your name listed as an author.
The most subtle point: the very people who truly want the model weights are the ones who are being bypassed.
Researchers working on AI have been demanding: give us the weights, give us the training data—because only then can we independently evaluate model behavior, verify reproducibility, and conduct security audits.
In response, OpenAI said that the weights cannot be provided to prevent misuse.
OpenAI has opened up access rights but tightly guards the model itself; the underlying system remains a black box.
Anthropic has been there all along, just with a different approach.
As early as May 2025, Anthropic launched the AI for Science initiative:
Up to $20,000 in API credits for researchers at universities and nonprofit organizations, valid for six months, with a focus on biology and life sciences.
But what the two companies are offering are not the same thing.
Anthropic provides API credits, which can only be used via API access; you cannot use the Claude web interface or access non-public experimental models. In short, it’s a compute voucher.
OpenAI provides a complete workspace, Pro-level limits, and team seats for five people.
One provides AI tools; the other provides an AI working environment. The latter is deeply tied to user habits.
However, the two share one surprising similarity: neither opens up weightings.
The scientists who truly study "AI itself" are precisely the ones who want the weights and training data.
Because only by obtaining these can they independently evaluate model behavior, reproduce results, and conduct security audits.
This plan targets students in mathematics, physics, chemistry, biology, and engineering, precisely avoiding the core needs of this group.
OpenAI and Anthropic stated that restricting access to weights helps prevent misuse.
Leaving these aside, for a researcher who has just received a one-year free quota, this tool benefit is real and substantial.
But when the free period ends, will you be able to leave?
Reference materials:
https://x.com/OpenAI/status/2082516370949062989
https://openai.com/index/chatgpt-for-academic-researchers/
This article is from the WeChat public account "New Intelligence Yuan," authored by ASI Revelation.