OpenAI's GPT-5 Accelerates Math and Science Research

What happened

OpenAI announced that its GPT-5 model is accelerating research in mathematics, biology, and physics. The model, launched on August 7, 2025, delivers state-of-the-art performance across scientific domains, achieving 94.6% on AIME 2025 mathematics benchmarks without tools. GPT-5 is now available to all ChatGPT users, with enhanced capabilities for complex problem-solving across multiple scientific disciplines.

Why it matters

• GPT-5 represents a significant leap in AI capability for research-intensive fields, potentially accelerating discovery timelines in mathematics, biology, and physics.
• The model's reduced hallucination rate—approximately 45% fewer factual errors than GPT-4o—makes it more reliable for scientific applications where accuracy is critical.
• AI groups are racing to develop advanced models, and GPT-5's demonstrated performance positions OpenAI as a leader in frontier AI development.

Key facts

1. 1. GPT-5 achieved 94.6% performance on AIME 2025 mathematics benchmarks without tools.
2. 2. GPT-5 launched on August 7, 2025, across ChatGPT, the OpenAI API, and GitHub Models Playground.
3. 3. GPT-5's responses are approximately 45% less likely to contain factual errors than GPT-4o when web search is enabled.
4. 4. GPT-5 performs better than OpenAI o3 with 50-80% fewer output tokens across visual reasoning, coding, and scientific problem-solving tasks.
5. 5. The model is available to all users, with Plus subscribers receiving more usage and Pro subscribers accessing GPT-5 Pro for extended reasoning.

Key Voices

• Not stated in the article.