OpenAI has announced the launch of its new series of AI models, codenamed "Strawberry," which are designed to enhance reasoning abilities and solve complex problems in fields such as science, coding, and mathematics.
The models, named o1 and o1-mini, are now available in ChatGPT and its API.
Key Takeaways
- OpenAI's new AI models, o1 and o1-mini, are designed to improve reasoning capabilities.
- The models can solve complex problems in science, coding, and mathematics.
- The o1 model scored 83% on the International Mathematics Olympiad qualifying exam.
- The models use a technique called "chain-of-thought" reasoning to break down complex problems.
- The o1-mini model is a cost-effective solution for developers, being 80% cheaper than o1-preview.
Introduction to the Strawberry Series
OpenAI has introduced a new series of AI models under the codename "Strawberry." These models are designed to spend more time processing answers to queries, enabling them to solve more challenging problems than previous models. The o1 and o1-mini models are the first in this series and are now available in ChatGPT and its API.
Enhanced Reasoning Capabilities
The o1 model has shown significant improvements in reasoning capabilities. It scored 83% on the qualifying exam for the International Mathematics Olympiad, compared to 13% for its predecessor, GPT-4o. The model also performed well in competitive programming questions and exceeded human PhD-level accuracy on a benchmark of science problems.
Chain-of-Thought Reasoning
One of the key techniques used in the new models is "chain-of-thought" reasoning. This involves breaking down complex problems into smaller logical steps, allowing the models to solve them more effectively. This technique has been automated in the new models, enabling them to break down problems on their own without user prompting.
Safety and Alignment
OpenAI has also focused on improving the safety and alignment of these new models. The o1-preview model scored 84 on a safety test, compared to 22 for GPT-4o. This was achieved by training the models to reason about safety rules in context, making them more effective at adhering to guidelines.
Cost-Effective Solutions for Developers
The o1-mini model offers a more efficient solution for developers, being 80% cheaper than the o1-preview model. It is particularly effective at coding and is designed to be a powerful, cost-effective model for applications that require reasoning but not broad world knowledge.
Availability and Future Plans
ChatGPT Plus and Team users can access the o1 models starting today, with Enterprise and Edu users gaining access next week. OpenAI also plans to bring o1-mini access to all free users of ChatGPT. Future updates will include additional features such as browsing, file and image uploading, and more.
OpenAI continues to develop and release models in its GPT series, alongside the new OpenAI o1 series, aiming to advance AI capabilities and solve complex problems more effectively.
Sources
- OpenAI to launch models with ‘reasoning’ abilities that are ‘much like a person’ | Artificial intelligence (AI) | The Guardian, The Guardian.
- OpenAI launches new series of AI models with 'reasoning' abilities | Reuters, Reuters.
- OpenAI releases new o1 reasoning model - The Verge, The Verge.
- OpenAI plans to release 'Strawberry' for ChatGPT in two weeks, Information reports | Reuters, Reuters.
- Introducing OpenAI o1 | OpenAI, OpenAI.
