OpenAI Releases GPT-4o Mini: Smaller, Faster, Lower Cost

2024-07-19

OpenAI has officially launched a streamlined and cost-effective version of its flagship AI model GPT-4o, called GPT-4o mini. The aim is to strike a perfect balance between functionality and cost, and further expand the accessibility of its cutting-edge AI tools. In the latest benchmark test for large-scale multitask language understanding (MMLU), GPT-4o mini has demonstrated impressive capabilities, achieving a high score of 82%. This score not only surpasses its predecessor GPT-3.5 Turbo's 70%, but also outperforms many competitors including Claude 3 Haiku (75.2%) and Gemini 1.5 Flash (78.9%). However, compared to the top-notch GPT-4o, GPT-4o mini still strives to reach its target score of 88.7%. In terms of technical specifications, GPT-4o mini is equipped with a 128K context window and comes at an affordable price: only $0.15 per million input tokens and $0.60 per million output tokens. In comparison, although Claude 3 Haiku offers a broader 200K context window, its cost is relatively higher at $0.25 per million input tokens and $1.25 per million output tokens. It is worth noting that although GPT-4o is designed as a comprehensive multimodal model, it currently focuses primarily on text and image processing and output. However, OpenAI has revealed plans to gradually introduce support for videos and audio, further expanding its capabilities. Starting today, GPT-4o mini is available to all free ChatGPT users, ChatGPT Plus subscribers, and team users. Enterprise users will have access in the coming week. This change means that GPT-3.5 Turbo will gradually fade out from the ChatGPT interface, but developers can still access GPT-3.5 through the API. The market response has been enthusiastic, with many companies quickly incorporating GPT-4o mini into their operations. For example, email service innovator Superhuman is using it to generate efficient replies, while financial services newcomer Ramp is leveraging the model to accurately extract key information from tedious receipts. To ensure safe usage, OpenAI has introduced a new security mechanism called "instruction-level" for GPT-4o mini. This mechanism sets instruction priorities to reduce potential misuse and enhance effective control over AI behavior. The release of GPT-4o mini is another important move by OpenAI to consolidate its market leadership and actively promote the widespread application of AI technology. By providing a compact and cost-effective model option, OpenAI aims to attract developers who may be considering switching to other competitors like Google's Gemini 1.5 Flash or Anthropic's Claude 3 Haiku, and together drive the popularization and development of AI technology.