OpenAI on Thursday announced GPT-4o mini, a simplified version of its GPT-4o large-scale language model, which the company says is more accurate than GPT-4 and significantly cheaper for developers to use than GPT-3.5 Turbo, facilitating the development of applications that make widespread use of AI models.
Additionally, there are two ways to build a custom AI assistant using GPT-4o, one of which is free.
The startup is touting its new AI model as “the most cost-effective small model on the market,” but as with most OpenAI releases, technical details about GPT-4o mini (such as the number of parameters) have not been made public, so it's unclear what “small” means in this case.
(An “AI model” is a part of an AI program that contains a number of neural net parameters and activation functions that are key elements of the AI program's functionality.)
The GPT-4o mini is priced at 15 cents per million input tokens and 60 cents per million output tokens, “an order of magnitude cheaper than previous state-of-the-art models and more than 60% cheaper than GPT-3.5 Turbo,” OpenAI wrote in a blog post emailed to ZDNET.
I also ran GPT-4o through a coding test and it passed all but one odd result.
The startup says the cost savings will help it develop applications that are sensitive to activity levels.
For example, applications that need to make multiple application programming interface (API) calls, applications that use a larger “context window” to retrieve materials (for example, retrieving an entire code base when developing an app), and applications that need to interact frequently with users, such as helpdesk support bots, would benefit from a lower cost per transaction, OpenAI said.
The model outperforms the standard GPT-4 model when used as a chatbot, based on crowdsourced testing from the Lmsys leaderboard, and “outperforms GPT-3.5 Turbo and other smaller models in academic benchmarks in both text intelligence and multimodal reasoning,” and supports the same number of languages as the standard GPT-4o model.
In common AI benchmarks such as the “MMU” test for “Multimodal Inference,” the GPT-4o mini-model outperforms several other models and approaches the performance of the standard GPT-4o.
In common AI benchmarks such as the “MMU” test for “Multimodal Inference,” the GPT-4o mini-model outperforms several other models and approaches the performance of the standard GPT-4o.
Open AI
Also, How to Digitize Your Handwritten Notes for Free with ChatGPT
The new model is immediately available to developers via the Assistant API, Chat Completion API, and Batch API, and can be used instead of GPT-3.5 Turbo on ChatGPT's Free, Plus, and Team accounts.
GPT-4o mini currently only offers support for text and images, with audio and video to be added at a yet to be determined date. GPT-4o mini's context window is 128,000 tokens, and training data is up to date until October 2023.
