OpenAI launches dedicated team to position AI as scientific collaborator across research fields

OpenAI Establishes Dedicated AI in Science Team

OpenAI has launched a new division called OpenAI for Science, marking an explicit push to position its large language models as partners in scientific research. Led by Kevin Weil, a vice president who joined the company as chief product officer after stints at Twitter and Instagram, the team was announced in October 2025 and aims to explore how AI tools for researchers can support work across mathematics, physics, chemistry, and biology 1

1

. Weil, who abandoned a particle physics PhD at Stanford for Silicon Valley, frames the initiative as central to OpenAI's mission of building artificial general intelligence that benefits humanity. "Maybe the biggest, most positive impact we're going to see from AGI will actually be from its ability to accelerate scientific research," he told MIT Technology Review 1

1

.

Scientists Send Millions of Messages Weekly

Internal analysis of anonymized ChatGPT conversations from January to December 2025 reveals substantial adoption among researchers. Average weekly message counts on advanced science and mathematics topics grew nearly 47% year-over-year, climbing from 5.7 million to approximately 8.4 million messages 2

2

4

4

. As of January 2026, nearly 1.3 million weekly users discuss advanced topics in scientific research with the AI, spanning graduate and research-level work 3

3

. Computer science, data science, and AI represent the most common domains, though engagement extends to computational chemistry, particle physics, and structural equation models 3

3

. Kevin Weil emphasized that "more researchers are using advanced reasoning systems to make progress on open problems, interpret complex data, and iterate faster in experimental work" 3

3

.

GPT-5 Achieves Breakthrough Performance on Research Benchmarks

The company attributes growing adoption to significant improvements in GPT-5's capabilities for complex problem-solving. On GPQA, an industry benchmark with over 400 multiple-choice questions testing PhD-level knowledge in biology, physics, and chemistry, GPT-5.2 reportedly scores 92% compared to GPT-4's 39% and a human-expert baseline of around 70% 1

1

. The model achieved gold-level results at the 2025 International Mathematical Olympiad and demonstrated partial success on the FrontierMath benchmark 2

2

. GPT-5.2 models can sustain long reasoning chains, verify results independently, and operate with formal proof systems like Lean 2

2

. OpenAI claims the models contributed to solutions connected to open Erdős problems, with human mathematicians confirming the results 2

2

.

How Researchers Integrate AI Into Research Workflows

Most scientists and engineers use ChatGPT for writing and communication tasks, while a smaller share employ it for rigorous analysis and calculations 3

3

. Researchers increasingly turn to the system as a scientific collaborator for routine, time-consuming activities including coding, literature synthesis, data interpretation, simulation support, and experiment planning 4

4

. In chemistry and biology, hybrid approaches pair general-purpose large language models with specialized tools such as graph neural networks and protein structure predictors 2

2

. Physics laboratories reportedly use AI to integrate simulations, experimental logs, documentation, and control systems while supporting theoretical exploration 2

2

. OpenAI cites case studies where AI shortened protein design timelines from years to months at RetroBioSciences 2

2

.

Playing Catch-Up With Google DeepMind

Despite the momentum, OpenAI enters a space where Google DeepMind has operated for years with groundbreaking scientific models such as AlphaFold and AlphaEvolve 1

1

. When Demis Hassabis, Google DeepMind's CEO and cofounder, discussed his firm's AI-for-science team in 2023, he stated: "This is the reason I started DeepMind ... In fact, it's why I've worked my whole career in AI" 1

1

. OpenAI's timing reflects recent advances in reasoning systems that have elevated AI capabilities beyond SAT-level performance to graduate-level work 1

1

. While the models do not generate entirely new mathematical theories, they recombine known ideas and identify connections across fields, which speeds up formal verification and scientific discovery 2

2

.

Questions About Validation and Long-Term Impact

Independent validation of OpenAI's reported gains remains limited 2

2

. Questions persist about how well these results hold up over time, how broadly they apply, and whether the reported gains translate into lasting scientific advances 2

2

. OpenAI argues that scientific progress supports medicine, energy systems, and public safety, yet research often advances slowly and requires substantial labor, with projects such as drug development taking more than a decade 2

2

. The company is urging policymakers to enhance science and research uses of AI, including scaling AI skilling, opening up data and frontier AI access to more people, and modernizing AI infrastructure 3

3

. For researchers watching this space, the key question is whether AI can move from handling routine tasks to contributing genuine insights that reshape how scientific discovery unfolds.