OpenAI on Thursday unveiled a stripped-down version of its GPT-4o large language model, GPT-4o mini, which it said has better accuracy than GPT-4 on tasks, and costs dramatically less than GPT-3.5 “Turbo” when used by developers, which it said can boost the construction of applications that use the AI model extensively.
The company touts the new AI model as “the most cost-efficient small model in the market,” although, as with most OpenAI releases, no technical details are available about GPT-4o mini, such as the number of parameters, hence, it’s unclear what “small” means in this case.
(An “AI model” is the part of an AI program that contains numerous neural net parameters and activation functions that are the key elements for how an AI program functions.)
Also: How to use ChatGPT to create an app
GPT-4o mini “is priced at 15 cents per million input tokens and 60 cents per million output tokens, an order of magnitude more affordable than previous frontier models and more than 60% cheaper than GPT-3.5 Turbo,” said OpenAI in a blog post emailed to ZDNET.
That reduction in cost, said the company, will aid the development of applications that are affected by volume of activity.
For example, applications that must make multiple API (application programming interface) calls, or that use larger “context windows” to retrieve materials (say, to retrieve an entire code-base when developing an app), or that have to interact frequently with the end user, such as a help desk support bot, will benefit from the reduction in per-transaction cost, said OpenAI.
The model, says OpenAI, outperforms the standard GPT-4 model when used as a chatbot, based on crowd-sourced tests by the Lmsys leaderboard. It also “surpasses GPT-3.5 Turbo and other small models on academic benchmarks across both textual intelligence and multimodal reasoning,” and supports as many languages as the standard GPT-4o model.
The new model is available immediately to developers via the Assistants API, Chat Completions API, and Batch API, and can be used instead of GPT-3.5 Turbo in ChatGPT’s free, plus, and team accounts.
The model offers only text and image support at the moment, with audio and video to be added at an unspecified date. The GPT-4o mini context window is 128,000 tokens, and its training data is current through October of 2023.