OpenAI Launches New Flagship Model GPT-4o Mini for Free Use in ChatGPT
OpenAI has unveiled its new flagship model, the GPT-4o Mini, an iteration of GPT-4o, today. This model boasts lower resource consumption and operational costs, enabling developers to integrate the model to provide AI services to a broader audience.
For developers, the launch of the GPT-4o Mini is significant news. Its usage cost is set at 0.15 per million tokens for input 0.60 per million tokens for output. This pricing is substantially lower than that of GPT-3.5 Turbo.
With its cost-effectiveness, the Mini version, built on the GPT-4o architecture, exceeds the capabilities of GPT-3.5 Turbo. Developers now have access to better results and lower operational costs with GPT-4o Mini.
For the general public, there's good news too. Users can now access the GPT-4o Mini model for free within ChatGPT. OpenAI announced that, starting today, users of the free version of ChatGPT, ChatGPT Plus subscribers, and ChatGPT Teams can utilize the GPT-4o Mini.
Starting next week, ChatGPT Enterprise subscribers will also be able to use the GPT-4o Mini model via the ChatGPT client or web version.
An additional benefit for ChatGPT free users is the seamless transition to GPT-4o Mini once the quota for the GPT-4o model is reached. This allows for uninterrupted conversations, unlike before when reaching the quota would halt interactions until the quota was restored.
The ability to lower API prices and serve ChatGPT free users is due to the reduced resource consumption of this scaled-down model. Not every AI task requires the full capabilities of full-size models like GPT, Claude, or Gemini.
Thus, many AI developers launch smaller models that can perform simple and repetitive tasks more quickly, economically, and efficiently, significantly reducing the overall operational costs and providing more users with free services or higher free quotas.
In the MMLU (Massive Multitask Language Understanding) inference benchmark, the GPT-4o Mini scored 82%, outperforming Google Gemini 1.5 Flash by 3% and Anthropic Claude 3 Haiku by 7%. For most users, this capability is more than sufficient for everyday use.
Lastly, OpenAI states that the GPT-4o Mini offers the same context window size as GPT-4o, which is 128K tokens, with the knowledge cutoff date also being the same, October 2023. Initially, the API will provide text and visual functionalities, with plans to include video and audio capabilities in the future.