OpenAI Announcing the launch of the latest lightweight generative model GPT-4o mini. Compared to the flagship GPT-4o model, the GPT-4o mini consumes fewer resources and costs less, making it easy for developers to apply AI technology to a wider range of products.
This is a major upgrade for developers and apps, and extends the ChatGPT The free version is functional and less restrictive.
Starting this Thursday, ChatGPT Free, Plus and Team Edition users can use the GPT-4o mini via the ChatGPT webpage and app, while enterprise users will have access to the GPT-4o mini starting next week.
Meanwhile, the GPT-4o mini will replace the existing GPT-3.5 Turbo model for all end users.
If developers don't want to switch to 4o mini yet, they can still use the old model through the API. openAI says they will eventually phase out the old model, but no specific date has been set yet.
Since May, the GPT-4o model has been available for free ChatGPT accounts, but you can only use GPT-4o a limited number of times in a three-hour period.With this update, there is still a limit on the number of times you can use GPT-4o, but when you reach the limit, the model automatically switches to the GPT-4o mini instead of GPT-3.5.
According to Artificial Analysis, OpenAI's latest AI model scored 821 TP3T on the MMLU inference benchmark, 31 TP3T higher than Gemini 1.5 Flash and 71 TP3T higher than Claude 3 Haiku.
In addition, OpenAI claims that the GPT-4o mini costs 60% less to run than the GPT-3.5 Turbo.The GPT-4o mini is priced at 15 cents per million input tokens and 60 cents per million output tokens.OpenAI calls the GPT-4o mini "the most powerful and cost-effective small model available today! ".
How are these cost savings realized? In reality, not every AI task requires a powerful large-scale model like GPT, Claude, or Gemini. Like killing a chicken with a bull's-eye, using large language models (LLMs) for simple but numerous tasks is a trivial waste of both money and computational resources.
This is where small language models like Google's Gemini 1.5 Flash, Meta's Llama 3 8b or Anthropic's Claude 3 Haiku come into play. They can perform these simple, repetitive tasks faster and more cost-effectively.
The GPT-4o mini has the same context window as the flagship GPT-4o model, with 128,000 tokens (equivalent to the contents of a book) and a knowledge deadline of October 2023, according to OpenAI. The model API currently only provides text and visual features, but video and audio features will be supported in the future.