OpenAI has unveiled a new addition to its lineup of language models: the GPT-4o Mini. This smaller, more affordable model is designed to spur AI development by making advanced capabilities more accessible to developers.
Why It Matters
The GPT-4o Mini, touted as smarter and cheaper than the earlier GPT-3.5 Turbo model, aims to significantly expand the range of applications built with AI. OpenAI hopes this new model will democratize AI development by lowering the barriers to entry.
Key Features
Both free and paid ChatGPT users can now access GPT-4o Mini, replacing GPT-3.5 which was released in November 2022. The new model supports text and vision in the OpenAI API, with future updates expected to include text, image, video, and audio inputs and outputs.
Enterprise users will gain access to GPT-4o Mini starting the week of July 22. OpenAI highlights that the model excels in mathematical reasoning and coding, and has also shown proficiency in tasks requiring reasoning. Companies like financial tech startup Ramp and email app Superhuman have already tested GPT-4o Mini for extracting data from files and generating email responses.
Technical Specifications
GPT-4o Mini has a context window of 128,000 tokens, matching that of the larger GPT-4o and far exceeding GPT-3.5 Turbo's 16,000 tokens. The model is priced at 15 cents per million input tokens and 60 cents per million output tokens, making it a cost-effective option for developers.
Impact and Future Vision
OpenAI envisions a future where AI models are seamlessly integrated into every app and website. The launch of GPT-4o Mini is a step towards that vision, offering developers the tools to build and scale powerful AI applications more efficiently and affordably.
GPT-4o Mini maintains the same safety parameters as GPT-4o and introduces a new safety technique called instruction hierarchy, which prioritizes prompts from developers over third parties to reduce vulnerability to external threats.
With GPT-4o Mini, OpenAI is paving the way for a new era of AI development, promising smarter, cheaper, and more accessible technology for a wide range of applications.