OpenAI has released its most cost-efficient miniature AI model, GPT-4o Mini, which is set to expand the scope of AI applications due to its affordable price and powerful capabilities. This model is substantially more cost-effective compared to its predecessors, such as GPT-3.5 Turbo, and is priced at 15 cents per million input tokens and 60 cents per million output tokens.
Interestingly, GPT-4o Mini has already excelled in performance against other mini AI models in various benchmarks such as reasoning tasks, math and coding proficiency, as well as multimodal reasoning. It scored 82% on the MMLU benchmark, outscoring Gemini Flash and Claude Haiku in reasoning tasks involving text and vision. It also recorded impressive scores of 87% on MGSM and 87.2% on HumanEval, indicating strong proficiency in mathematical reasoning and coding tasks. Furthermore, it scored 59.4% on the MMMU benchmark, demonstrating better performance than Gemini Flash and Claude Haiku in multimodal reasoning evaluation.
The model has a context window of 128k tokens and is compatible with text and vision inputs and outputs, attesting to its versatility. OpenAI is planning future updates to extend support for image, video, and audio inputs and outputs. The model excels in applications that require chaining or parallelizing multiple model calls, handling large volumes of context, or providing fast, real-time text responses.
OpenAI has also incorporated significant safety measures into GPT-4o Mini. The organization has filtered out harmful content and utilized Reinforcement Learning with Human Feedback (RLHF) to enhance its safety. The model also employs an instruction hierarchy method to resist jailbreaks and excessive prompts, ensuring it’s appropriate for large-scale application use.
The GPT-4o Mini is readily available to developers through various APIs, enabling immediate access for Free, Plus, and Team users of ChatGPT, with enterprise users slated to gain access in the following week. As part of a drive by OpenAI to decrease costs while improving model capabilities, fine-tuning for the GPT-4o Mini will soon be available to further improve its usability and accessibility.
In summary, OpenAI’s GPT-4o Mini has set a new benchmark in cost-effective and high-performing AI models, making AI much more accessible to developers and businesses. This miniature model supports different applications and promises to make significant contributions to the future trajectory of AI.
