OpenAI Unveils GPT-4o Mini, Bringing Affordable AI to a Wider Audience

OpenAI has introduced GPT-4o mini, its most cost-efficient small model to date, aiming to make advanced artificial intelligence accessible to a broader audience. This launch is expected to significantly expand the range of AI applications by offering a more affordable option for developers and businesses.

GPT-4o mini is designed to handle various tasks, including video, audio, images, and text, making it a versatile tool for multiple industries. This model achieves an impressive 82% on the MMLU benchmark and outperforms GPT-3.5 Turbo on the LMSYS leaderboard for chat preferences. At a price of 15 cents per million input tokens and 60 cents per million output tokens, it is more than 60% cheaper than its predecessor, GPT-3.5 Turbo.

GPT-4o mini's affordability opens up new possibilities for applications that require chaining or parallelizing multiple model calls, processing large volumes of context, or delivering real-time text responses. This makes it an ideal choice for customer support chatbots and similar applications.

Currently, GPT-4o mini supports text and vision inputs through the API, with plans to extend support to image, video, and audio inputs and outputs in the future. The model features a 128K token context window and can handle up to 16K output tokens per request, with knowledge updated up to October 2023. Additionally, it benefits from an improved tokenizer that enhances its ability to handle non-English text cost-effectively.

Benchmark tests show that GPT-4o mini surpasses other small models in various areas. It scores higher in textual intelligence and multimodal reasoning tasks and demonstrates strong performance in function calls, which is crucial for building applications that interact with external systems. Specifically, it scored 82.0% on the MMLU benchmark, outpacing Gemini Flash and Claude Haiku. In mathematical reasoning and coding tasks, it achieved 87.0% on MGSM and 87.2% on HumanEval, outperforming previous models.

OpenAI has emphasized the importance of safety in its models. GPT-4o mini incorporates robust safety measures, filtering out undesirable content during pre-training and aligning model behaviour with OpenAI’s policies through reinforcement learning with human feedback (RLHF). The model also employs an instruction hierarchy method to resist jailbreaks and prompt injections, enhancing its reliability for large-scale applications.

GPT-4o mini is now available through the Assistants API, Chat Completions API, and Batch API. Developers can access it for 15 cents per million input tokens and 60 cents per million output tokens. Additionally, ChatGPT users, including Free, Plus, and Team members, can use GPT-4o mini starting today, with Enterprise access becoming available next week.

This launch marks a significant step in OpenAI’s mission to make artificial intelligence more accessible and integrated into everyday digital experiences. As the cost of AI continues to decrease, the capabilities of models like GPT-4o mini are expected to become increasingly embedded in applications across various sectors, driving innovation and efficiency.