AI BriefingOpenAI Unveils New Reasoning Model o1
Key point
OpenAI has unveiled o1, a new series of AI models with enhanced reasoning capabilities designed to solve complex math, science, and coding problems.
Details
OpenAI has unveiled o1, a new series of AI models designed to spend more time thinking before responding. These models go through a reasoning process to solve complex problems in science, coding, and mathematics, and they have the ability to revise their own strategies and recognize mistakes.
Key performance metrics are as follows:
- Math: Achieved an 83% score on International Mathematical Olympiad (IMO) qualifying problems, compared to GPT-4o's 13%
- Coding: Reached the top 89% percentile in Codeforces competitions
- Science: Demonstrated PhD-level performance on physics, chemistry, and biology benchmarks
The model is offered in two versions:
- o1-preview: A model optimized for complex tasks that require advanced reasoning
- o1-mini: A fast and affordable model specialized for coding, costing 80% less than o1-preview
In terms of safety, the model's reasoning capabilities were leveraged to improve its ability to comply with guidelines. As a result, it scored 84 points in jailbreaking tests, far surpassing GPT-4o's 22 points, strengthening its security.
Currently, ChatGPT Plus and Team users can directly select and use o1-preview and o1-mini through the model picker.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.