On Thursday, OpenAI announced and immediately put into operation the slimmed-down, cheaper version of the GPT-4o large language model introduced in the spring, the GPT-4o mini.
According to the startup, which is now estimated to be worth roughly 80 billion dollars thanks to the AI boom it catalyzed, supported by Microsoft among others, the GPT-4o mini is currently the cheapest and most knowledgeable small model available.
One of the most important aspects of the announcement may be the pricing of the new mini model, which allows application developers to build AI-based applications and services up to 60% cheaper than using the currently widespread GPT-3.5 Turbo.
An important change is that from yesterday’s premiere, ChatGPT users (both free and subscription plans) can use the new model instead of the retiring GPT-3.5 Turbo, but the older model will remain available to developers at the API level for the time being undefined for a while.
Google and Anthropic have already introduced a similarly simplified model (Gemini 1.5 Flash and Claude 3 Haiku), but according to OpenAI, the GPT-4o mini performs better than all competitors in independent benchmarks (mainly the MMLU exam).
Source: www.hwsw.hu
