OpenAI has introduced GPT-4o mini, its most cost-efficient small language model, designed to expand AI accessibility by offering high performance at significantly reduced prices.
GPT-4o mini is priced at 15 cents per million input tokens and 60 cents per million output tokens, making it 60% cheaper than GPT-3.5 Turbo. Despite its lower cost, the model demonstrates high performance, scoring 82% on MMLU (Massive Multitask Language Understanding) and outperforming GPT-41 on chat preferences in the LMSYS leaderboard.
The model supports text and vision in the API, with a 128K token context window and up to 16K output tokens per request. It excels in reasoning tasks, math and coding proficiency, and multimodal reasoning, outperforming other small models on various benchmarks.
OpenAI has incorporated the same safety mitigations as GPT-4o and introduced new techniques like the instruction hierarchy method, to improve reliability and safety.
OpenAI emphasises that this release is part of their commitment to making AI more accessible and affordable, noting that the cost per token has dropped by 99% since the introduction of text-davinci-003 in 2022.