OpenAI Launches o1 Models Capable of Self-Fact-Checking Through Extended Reasoning
OpenAI introduces a new family of AI models called o1 that spend more time "thinking" to fact-check themselves and solve complex problems. The release currently faces high costs, strict rate limits, and missing features like web browsing.
OpenAI officially unveils its newest family of generative AI models, code-named Strawberry and now called OpenAI o1. The company releases two versions, o1-preview and o1-mini, to ChatGPT Plus and Team subscribers, with Enterprise and educational users gaining access early next week. Despite the excitement, the current chatbot experience is fairly barebones because the model lacks the ability to browse the web or analyze files, and its image-analyzing features remain disabled pending additional testing.
What sets o1 apart from previous models is its unique ability to "think" before responding. Trained with reinforcement learning, the system uses a private chain of thought to plan ahead and break down complex tasks over an extended period. This self-fact-checking mechanism allows o1 to reason through problems holistically, making it highly effective for complicated duties like detecting privileged legal emails or brainstorming detailed marketing strategies.
Accessing this advanced reasoning comes with significant limitations and high costs. Users face strict weekly message caps of 30 for o1-preview and 50 for o1-mini, while API developers pay a steep $15 per million input tokens and $60 per million output tokens. These prices represent a massive sixfold increase over the costs of using GPT-4o, though OpenAI promises to eventually bring the smaller o1-mini model to all free ChatGPT users at an unannounced future date.