OpenAI analysed more than 580,000 messages from 1,819 scientists enrolled in its ChatGPT for Academic Researchers program and published the results as “Science at the Frontier”. The model was used across the research workflow: 68% of participants used it to develop research ideas, and others for literature review, data analysis, code and drafting publications.

The headline saving is self-reported. Respondents estimated that the same work at comparable quality would have taken 1.9 times longer without AI, a 47% reduction, and 98% expect AI to speed up discovery in their field.

OpenAI says staff analysed request categories without reading the conversations. A participant estimate is not a measured comparison, so the number describes perceived speed, not a benchmark of output.

Science at the Frontier · Full report (PDF) · Source channel