- OpenAI claims that 8.4 million weekly messages are sent about advanced science and mathematics
- GPT-5.2 models can follow long chains of reasoning and independently verify results
- AI speeds up routine research tasks like coding, literature review and experiment planning
OpenAI wants users to treat ChatGPT as a research partner, with new research claiming that nearly 8.4 million messages are sent each week focused on advanced science and math topics, generated by around 1.3 million users worldwide.
OpenAI highlights that this use has grown nearly 50% over the past year, suggesting that the system is moving beyond occasional experimentation and into mainstream research workflows.
These users reportedly engage in work comparable to graduate-level study or active research across mathematics, physics, chemistry, biology, and engineering.
Application scale and research integration
Mathematics receives special attention in the report. GPT-5.2 models are said to maintain long chains of reasoning, control their own work, and operate with formal evidence systems such as Lean.
OpenAI claims the models achieved gold-level results at the 2025 International Mathematical Olympiad and demonstrated partial success on the FrontierMath benchmark.
The report also states that the models have contributed to solutions associated with open Erdő’s problems, with human mathematicians confirming the results.
Although the models do not generate entirely new mathematical theories, they recombine known ideas and identify connections across fields, speeding up formal verification and evidence discovery.
Similar patterns appear in other scientific fields. On candidate-level benchmarks such as GPQA, GPT-5.2 reportedly exceeds 92% accuracy without external tools.
Physics laboratories reportedly use artificial intelligence to integrate simulations, experimental logs, documentation and control systems, while also supporting theoretical exploration.
In chemistry and biology, hybrid approaches pair general language models with specialized tools such as graphical neural networks and protein structure predictors.
These combinations aim to improve reliability while keeping human oversight central to decision-making.
The report places this development in a wider context. Scientific advances support medicine, energy systems and public safety, but research is often slow and requires significant manpower.
A small portion of the global population produces most fundamental discoveries, while projects such as drug development can take more than a decade.
OpenAI claims that researchers are increasingly using AI tools to handle routine, time-consuming tasks, including coding, literature review, data analysis, simulation support, and experiment planning.
It cites case studies ranging from faster mathematical proofs to protein design with RetroBioSciences, where AI reportedly shortened timelines from years to months.
Although the report presents notable usage figures and benchmark results, independent validation remains limited.
Questions remain about how well these findings hold up over time, how broadly they apply, and whether the reported gains translate into lasting scientific advances.
These usage numbers and benchmark scores stand out, but independent validation is still limited.
Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews and opinions in your feeds. Be sure to click the Follow button!
And of course you can too follow TechRadar on TikTok for news, reviews, video unboxings, and get regular updates from us on WhatsApp also.
