OpenAI has launched a groundbreaking series of AI models designed to tackle complex problems in science, coding, and mathematics. Introducing OpenAI o1, a new model preview that promises a significant advancement in AI’s reasoning capabilities. Available starting from September 12th, this series aims to solve more challenging tasks than its predecessors by “thinking” more before responding.
How OpenAI o1 Works
This new model is trained to spend extra time reasoning through problems, much like a human. It refines its thinking process, tries different strategies, and learns from its mistakes. This approach has led to remarkable improvements in handling complex tasks. In tests, the model performed on par with PhD students on advanced physics, chemistry, and biology challenges. Its capabilities in coding and math are equally impressive—scoring 83% on a qualifying exam for the International Mathematics Olympiad and reaching the 89th percentile in Codeforces competitions.
A Focus on Safety
OpenAI places a high priority on the safety and ethical use of its models. OpenAI o1 comes with a new safety training approach, enabling the model to reason about safety rules effectively in various contexts. During rigorous testing to prevent “jailbreaking” attempts (efforts to bypass safety rules), OpenAI o1 scored 84 out of 100, significantly outperforming previous models.
To match the enhanced capabilities of the new model, OpenAI has strengthened its internal governance and collaborates closely with federal government agencies. The company has also formalized agreements with the U.S. and U.K. AI Safety Institutes, providing early access to research versions of the model for evaluation.
Related Story: OpenAI Surpasses 1 Million Paid Users and Prepares for New Advanced Models
Who Can Benefit from this latest model?
OpenAI o1 is particularly valuable for professionals and researchers dealing with complex problems in science, coding, and mathematics. It can be used in various scenarios, such as:
- Healthcare Researchers: Assisting in annotating cell sequencing data.
- Physicists: Generating complex mathematical formulas for quantum optics research.
- Software Developers: Building and executing multi-step workflows.
OpenAI o1-Mini: An Affordable Option
Alongside OpenAI o1, the company is also introducing OpenAI o1-mini—a more efficient, cost-effective model designed primarily for coding tasks. OpenAI o1-mini is 80% cheaper than o1-preview, making it a powerful option for applications requiring focused reasoning without extensive world knowledge.
Accessing OpenAI o1
ChatGPT Plus and Team users can access OpenAI o1 models immediately, selecting either o1-preview or o1-mini manually from the model picker. At launch, weekly rate limits are set at 30 messages for o1-preview and 50 for o1-mini, with plans to increase these limits over time.
ChatGPT Enterprise and Edu users will gain access to these models starting next week. Developers meeting API usage tier 5 requirements can also begin prototyping with OpenAI o1 models through the API, which currently has a rate limit of 20 RPM.
OpenAI plans to extend access to o1-mini for all ChatGPT Free users soon, ensuring broader accessibility for coding and problem-solving tasks.
