OpenAI's new o3-mini model is designed to excel at reasoning, especially in the areas of mathematics, programming, and science. OpenAI says the o3-mini "pushes the boundaries of what mini models can achieve" and is 24% faster at responding than the previous o1-mini model. Independent testers say the o3-mini's answers are 56% more likely to be preferred than the o1-mini, and errors are reduced by 39%.
OpenAI releases o3-mini earlier than tech industry predicted
Like the o1-mini, the o3-mini allows users to track the model's thinking process through feedback. Users can also combine this reasoning with web searches, although this feature is still in development.
o3-mini is OpenAI's response to DeepSeek
The o3-mini launch comes amid fierce competition in the AI space, with China’s DeepSeek’s free R1 model attracting millions of users thanks to its speed of response, high accuracy, and much lower operating costs. OpenAI is trying to regain user attention and traffic.
Users of the free ChatGPT plan can access o3-mini by clicking the “Reason” button in the text input box. While OpenAI has not announced specific usage limits, it is likely to be similar to the current restrictions for GPT-4o, which is that users will receive a certain number of queries per hour.
For paid users, o3-mini can be selected from the model selector in the upper left corner. Plus or Team plan users will receive 150 o3-mini queries per day, while Pro plan users, which cost $200/month, will have unlimited access.
Additionally, paid ChatGPT users can also access the o3-mini-high model, which has similar reasoning capabilities but longer response times. This model promises improved performance if users don't mind waiting a few extra seconds.
OpenAI also stressed that o3-mini underwent rigorous safety assessments before its launch, which showed that the model "significantly outperforms" GPT-4o in handling unsafe scenarios and cracking attempts.
Source: https://thanhnien.vn/openai-phat-hanh-o3-mini-mien-phi-truoc-suc-ep-deepseek-185250202011922867.htm
Comment (0)