OpenAI has launched GPT-4o mini, a smaller and more cost-effective version of its generative AI model, GPT-4o. This new model promises improved performance and accessibility for developers and users alike. GPT-4o mini is available now for Free, Plus, and Team ChatGPT users via the web and app, with Enterprise subscribers gaining access next week. It will replace the current GPT-3.5 Turbo model for end-users, offering a significant upgrade in capabilities.
Developers can still access GPT-3.5 Turbo through the API, although OpenAI plans to eventually retire the older model. The introduction of GPT-4o mini enhances the free tier of ChatGPT by providing a fallback option when GPT-4o usage limits are reached. Previously, the fallback was GPT-3.5.
Performance and Cost Efficiency
According to Artificial Analysis, GPT-4o mini achieved an 82% score on the MMLU reasoning benchmark, surpassing Gemini 1.5 Flash and Claude 3 Haiku. While Google’s Gemini Ultra still holds the top spot at 90%, GPT-4o mini demonstrates impressive performance for its size. Furthermore, OpenAI asserts that GPT-4o mini is 60% cheaper to operate than GPT-3.5 Turbo, costing developers 15 cents per million input tokens and 60 cents per million output tokens. OpenAI touts GPT-4o mini as the most capable and cost-efficient small model currently available.
Smaller LLMs: Addressing Specific Needs
The need for smaller LLMs stems from the fact that not all AI tasks require the full power of larger models. Using a standard LLM for simple, high-volume tasks can be inefficient and costly. Smaller LLMs like GPT-4o mini, Gemini 1.5 Flash, Llama 3 8b, and Claude 3 Haiku are designed to handle these tasks with greater speed and cost-effectiveness.
Key Features and Future Developments
GPT-4o mini boasts a 128,000-token context window, matching the full-sized GPT-4o, and possesses the same knowledge cutoff date of October 2023. While the exact size of the new model remains undisclosed, the API currently supports text and vision capabilities, with video and audio functionalities planned for the future.
Following the Voice Mode Update
This announcement follows OpenAI’s recent update on its forthcoming Voice Mode for GPT-4o. A smaller alpha release is expected in late July, with a broader rollout planned for this fall.
This launch signifies OpenAI’s ongoing commitment to making advanced AI more accessible and affordable. With its enhanced performance and cost-effectiveness, GPT-4o mini is poised to empower developers and users in various applications.
