OpenAI has introduced GPT-4o mini, a smaller and less resource-intensive version of its GPT-4o model, aimed at making AI technology more accessible to developers and users. This new model is cheaper to operate than the standard GPT-4o, enabling developers to integrate AI into a wider range of products. The update also benefits free ChatGPT users, expanding capabilities and reducing limitations. GPT-4o mini is now accessible on the Free, Plus, and Team tiers of ChatGPT for both users and developers, with ChatGPT Enterprise users gaining access next week. Replacing the previous GPT-3.5 Turbo, GPT-4o mini is the default model for end users, while developers can continue utilizing GPT-3.5 Turbo through the API. OpenAI plans to retire the older model eventually but has not announced a specific date. GPT-4o has been available to free ChatGPT accounts since May, but limitations due to high demand have been in place. While those limitations remain for the full GPT-4o model, users will now be downgraded to GPT-4o mini instead of GPT-3.5 when they hit their limit, offering a significant advantage for free users. Artificial Analysis data reveals that GPT-4o mini achieved an 82% score on the MMLU reasoning benchmark, surpassing Gemini 1.5 Flash by 3% and Claude 3 Haiku by 7%. While Google’s top-of-the-line AI, Gemini Ultra, holds the highest benchmark score at 90%, GPT-4o mini’s performance is notable. Furthermore, OpenAI claims that GPT-4o mini is 60% cheaper to operate than GPT-3.5 Turbo, with developers paying 15 cents per million input tokens and 60 cents per million output tokens. OpenAI highlights GPT-4o mini as “the most capable and cost-efficient small model available today,” according to CNBC. The cost savings stem from the realization that not all AI tasks require the extensive capabilities of larger models like GPT, Claude, or Gemini. Utilizing a full-sized LLM for simple but high-volume tasks is inefficient, leading to wasted resources and unnecessary costs. Smaller LLMs, such as Google’s Gemini 1.5 Flash, Meta’s Llama 3 8b, and Anthropic’s Claude 3 Haiku, excel at handling these simple, repetitive tasks more efficiently. OpenAI states that GPT-4o mini maintains the same 128,000-token context window (roughly the size of a book) as the full-sized version, along with the same knowledge cutoff of October 2023. However, the exact size of the new model remains undisclosed. Currently, the model API provides text and vision capabilities, with plans for video and audio support in the future. This announcement comes shortly after OpenAI shared an update on its highly anticipated Voice Mode for GPT-4o. A smaller alpha release is scheduled for late July, with a wider rollout anticipated for fall.
OpenAI Unveils GPT-4o Mini: A Smaller, Cheaper, and More Accessible AI Model
