OpenAI launches new AI model - 'OpenAI o1'

OpenAI has just released a revolutionary new AI model, known internally as "Strawberry," which boasts advanced reasoning capabilities for tackling complex problems in math, science, and coding. This new model, named OpenAI o1-preview, represents a significant leap forward in artificial intelligence technology.

OpenAI o1-Preview

Starting September 12, OpenAI has launched a new series of reasoning models designed to tackle intricate problems more effectively than ever before. These models are crafted to think more deeply and carefully before providing answers, a feature that sets them apart from previous AI versions.

What Makes o1-Preview Special?

The OpenAI o1-preview model excels in complex reasoning tasks by analyzing problems similarly to how humans do. This enhanced thinking process allows it to explore various strategies, refine its approaches, and learn from mistakes. In testing, this model performed at a level comparable to PhD students in subjects like physics, chemistry, and biology. For instance, while previous models like GPT-4o solved only 13% of International Mathematics Olympiad problems correctly, the new reasoning model achieved an impressive 83%.

In coding, the o1-preview model has reached the 89th percentile in Codeforces competitions, showcasing its superior abilities in generating and debugging code. However, it's important to note that this early model version lacks some of the features present in GPT-4o, such as web browsing and file uploading. Despite this, its advancements in reasoning represent a significant upgrade in AI capabilities.

Advancements in Safety with o1-Preview

OpenAI has also introduced a new safety training approach alongside the o1-preview model. 

This approach leverages the model's reasoning abilities to better adhere to safety and alignment guidelines. For instance, in a challenging safety test, the o1-preview model scored 84, a notable improvement over GPT-4o's score of 22.

The development of these models is supported by rigorous testing and collaboration with AI safety institutes in the U.S. and U.K. This ensures that the new models meet high safety standards before and after their public release.

Who Can Benefit from o1-Preview?

The OpenAI o1-preview model is particularly beneficial for professionals and researchers working on complex problems in science, mathematics, and coding. Healthcare researchers, physicists, and developers can use this model to streamline tasks such as annotating cell sequencing data, generating advanced mathematical formulas, and executing multi-step workflows.

Introducing OpenAI o1-Mini

For those seeking a more cost-effective solution, OpenAI is also releasing the o1-mini model. This smaller, faster model offers similar reasoning capabilities but at 80% less cost than the o1-preview. It is designed for efficient coding tasks and is ideal for applications requiring reasoning without extensive world knowledge.

How to Access OpenAI o1 Model?

Starting today, ChatGPT Plus and Team users can access both o1-preview and o1-mini models through ChatGPT. Initial weekly rate limits are set at 30 messages for o1-preview and 50 for o1-mini. Access for ChatGPT Enterprise and Edu users will begin next week.

Developers can start using these models through the API, with a rate limit of 20 requests per minute. Note that some features, like function calling and streaming, are not yet available in the API. More information can be found in the API documentation.

What’s Next for OpenAI?

This release marks the beginning of a new era for reasoning models. Future updates will bring additional features such as browsing and file uploading. OpenAI is committed to continuing the development of both the GPT and o1 series to enhance AI capabilities further.