OpenAI is releasing a new AI model called o1 that can perform human-like reasoning tasks. This is the first of the many planned series of reasoning models that have been trained to answer more complex questions.
There is also the o1-mini, a smaller and cheaper version of the same. This is the same super-hyped Strawberry model.
This new model from OpenAI represents a step towards its broader goal of creating a human-like artificial intelligence. It does a better job at writing code and solving multistep problems than the previous AI models from OpenAI. However, this is expensive to use and on the slower side when compared to GPT-4o. To explain that it is at the very beginning of its complete form, OpenAI is calling o1 a “preview” of what is to come.
Starting from September 12th, ChatGPT Plus and Team users will get access to both o1-preview and o1-mini. The Enterprise and Edu users will get access to these models early next week. They are also saying that they plan to release the o1-mini for all free users of ChatGPT. However, they have not disclosed the timeline of release.
The API access to this o1-preview is $15 per one million input tokens, or chunks of text parsed by the model and $60 per one million output tokens. So, it is very expensive as of now.
The company is vague about the details of its training but it has confirmed that the new model has been trained using a completely new optimization algorithm and a new training dataset.
The previous models mimic patterns from their training data. The new o1 model is trained to solve problems on its own using a technique known as reinforcement learning, which teaches the system through rewards and penalties.
