OpenAI, led by Sam Altman, has rolled out the eagerly awaited o1-preview collection of AI models, marking a significant advancement in artificial intelligence, particularly for ChatGPT. These cutting-edge models excel in tackling intricate problems across various fields such as science, coding, and mathematics.
Under the leadership of Sam Altman, OpenAI has unleashed the latest AI models into ChatGPT and the API for early access, with plans for continuous enhancements. Expressing his enthusiasm for this launch, Altman took to X (formerly Twitter) to convey his pride in the team and the monumental effort involved, inviting users to enjoy the new models.
The o1-preview models bring a higher level of reasoning to the table, dedicating more time to processing information before formulating responses. This enhancement results in improved problem-solving capabilities. Initial tests have shown that the updated reasoning model performed comparably to PhD students in tasks related to physics, chemistry, and biology, and excelled in math and coding competitions.
In a notable example, the o1 model obtained an impressive 83% score in a qualifying exam for the International Mathematics Olympiad, surpassing GPT-4o’s 13%. Despite its advanced functionalities, the o1-preview model lacks some practical features found in GPT-4, such as web browsing and file uploading.
Evidently, OpenAI, with backing from Sam Altman, emphasizes the model’s prowess in handling complex, multi-step tasks, rendering it highly valuable for industries necessitating advanced problem-solving capabilities. Altman highlighted this evolution in AI performance, heralding the dawn of a new era involving AI capable of intricate, general-purpose reasoning.
To cater to developers seeking advanced coding abilities without extensive background knowledge, OpenAI also introduced a more compact and cost-effective version known as o1-mini. Priced 80% lower than the o1-preview, the o1-mini model offers a budget-friendly solution for developers.
Users of ChatGPT Plus and Team can now manually opt for o1-preview and o1-mini from the model selector. While the o1-preview model has a rate limit of 30 messages, the o1-mini model allows for 50 messages. Additionally, API users in the highest usage tier can now leverage these models.
Despite some features like function calling and streaming not being available, OpenAI has incorporated new security training measures focusing on safety. In security tests, the o1-preview outperformed GPT-4o, achieving a score of 84 out of 100 compared to GPT-4o’s 22.
OpenAI has bolstered its partnerships with AI safety institutes in the U.S. and U.K. to reinforce its safety protocols. Future plans for OpenAI include expanding o1-mini access for ChatGPT Free users and introducing new features to the o1 series, including a browsing functionality and file uploads.
Altman concluded the announcement by acknowledging the models’ imperfections while remaining optimistic about their promising prospects, highlighting the initiation of a groundbreaking paradigm. OpenAI is now aiming for a $150 billion valuation following its latest funding rounds.