OpenAI, supported by Microsoft, recently released a series of new artificial intelligence models designed to enhance reasoning capabilities for complex tasks. These new models are named o1 and o1-mini, where "o" stands for Orion. It is reported that these models can handle more complex scientific, programming, and mathematical problems compared to previous versions.
OpenAI stated that the o1 model has been trained through reinforcement learning and can simulate human thought processes by spending more time analyzing problems before making decisions. This approach allows the model to continually optimize its reasoning process, find more effective solutions, and learn to identify and correct mistakes.
It is worth mentioning that the usage cost of o1 has significantly increased. Developers calling o1-preview through the API will incur costs of $15 per million tokens for input and $60 for output. In contrast, the costs for GPT-4o are only $5 for input and $15 for output. As a more economical option, the o1-mini model focuses on the coding field and is 80% cheaper than o1-preview.
Currently, ChatGPT Plus users and team users already have access to the o1-preview and o1-mini models. Enterprise and educational users will be able to use them starting next week. OpenAI stated that in the future, they will gradually open access to the o1-mini model to all free users.
