Announced recentlyNew GPT-4o artificial intelligence modelAfter that, OpenAI launched its lightweight version model earlier andGPT-4o miniFor the name.
Compared to the MMLU (Measuring Massive Multitask Language Understanding) measurement tool, GPT-4o mini's multitask language understanding reached 82%. This is slightly lower than the 4% of the previously released GPT-88.7o, but much higher than the 3.5% of GPT-70 Turbo.
OpenAI said that compared to the previously launched smallest GPT-3.5 Turbo, developers will be able to reduce the cost of building services with GPT-4o mini by 60%. OpenAI will also replace GPT-4 Turbo with GPT-3.5o mini as the model behind the free version of ChatGPT, which will also correspond to faster response times.
However, at present, GPT-4o mini only supports text and image understanding and generation, but OpenAI stated that it will continue to add support for content such as sound and video in the future to compete with lightweight large-scale natural language models proposed by other companies.
Although GPT-3.5 Turbo is no longer used in the ChatGPT service, OpenAI still retains the option of accessing GPT-3.5 Turbo through the API for existing developers, and application services previously built with GPT-3.5 Turbo or GPT-3.5 will not be affected.
