OpenAI Introduces GPT-4o Mini, Its Most Cost-Efficient Small AI Model

What to know

OpenAI has introduced GPT-4o mini – a new small AI model that is faster and more affordable than its previous models.
GPT-4o mini outperforms other small AI models across several benchmarks, and will replace GPT-3.5 Turbo as OpenAI’s smallest model.
GPT-4o mini is available for developers as well as consumers on ChatGPT web and its mobile apps.

Two months after GPT-4o’s grand introduction, OpenAI has now unveiled a smaller, faster version of their latest AI model – GPT-4o mini.

The new model will replace GPT-3.5 Turbo which has hitherto been the go-to model for free ChatGPT users, especially when the server is overloaded and faster responses are called for.

As per GPT-4o mini’s ‘Model Evaluation Scores’, the small AI model gave a consistently strong performance compared to other models like Gemini Flash, Claude Haiku, and GPT-3.5 Turbo across several key benchmarks. These included reasoning tasks, math and coding, and multimodal reasoning. Only its elder cousin GPT-4o performed better.

GPT-4o mini is also much more affordable. Developers using the API will pay just 15 cents for 1M input tokens and 60 cents for 1M output tokens. Compared to GPT-3.5 Turbo, the new GPT-4o mini model is 60% cheaper while having superior textual intelligence and multimodal reasoning.

The model currently supports text and vision in the API with a context window of 128K tokens. However, support for text, image, video and audio inputs and outputs is yet to arrive. Knowledge cutoff for GPT-4o mini is also still October 2023, similar to GPT-4o.

In terms of AI models that can practically be used for everyday tasks (by consumers) and for the creation of apps (by developers), it’s usually the cheaper, smaller models that are more in demand. So, it’s no surprise that OpenAI has made GPT-4o mini available to developers as well as consumers via ChatGPT web and mobile app.

The evolution of small AI models has made them much more popular with developers than large models in general. Although OpenAI hasn’t revealed the exact size of GPT-4o mini, its comparisons with Claude Haiku, Gemini Flash, and Llama 3 8b put it in the same category as these small AI models.

OpenAI expects the release of GPT-4o mini will enable “developers to build and scale powerful AI applications more efficiently and affordably“.