OpenAI recently announced the release of its highly anticipated newest snack-sized generative model dubbed GPT-4o mini. This model is not only less resource-intensive but also more affordable to operate compared to the standard GPT-4o. It allows developers to effortlessly integrate the AI technology into a vast array of products. This is a significant upgrade for developers and apps, as it expands the capabilities and reduces limitations on the free version of ChatGPT. GPT-4o mini is now accessible to users on the Free, Plus, and Team tiers through the ChatGPT web and app starting today, and ChatGPT Enterprise subscribers will gain access next week. Additionally, it will gradually replace the company’s existing small model, GPT-3.5 Turbo, for end users from today.

The older model remains available to developers through the API if they don’t wish to switch immediately. The company plans to eventually retire the older model but hasn’t specified a date. GPT-4o has been available to free ChatGPT accounts since May, but there were limitations on demand. Now, when hitting the limit, users will be downgraded to GPT-4o mini rather than GPT-3.5. This is a great advantage for those who haven’t upgraded to ChatGPT Plus.

“We’re continuing to make advanced AI accessible to all with the launch of GPT-4o mini, now available in the API and rolling out in ChatGPT today.” said OpenAI (@OpenAI) on July 18, 2024.

According to data from Artificial Analysis, OpenAI’s newest AI model scored 82% on the MMLU reasoning benchmark, surpassing Gemini 1.5 Flash by 3% and Claude 3 Haiku by 7%. The highest MMLU benchmark to date was set by Google’s top-of-the-line AI, Gemini Ultra, with a score of 90%. Moreover, OpenAI claims that GPT-4o mini is 60% cheaper to operate than GPT-3.5 Turbo. Developers will pay 15 cents per million input tokens and 60 cents per million output tokens. It is touted as the most capable and cost-efficient small model available today, as per CNBC.

Where do these cost savings come from? Well, not every task that can be enhanced by AI requires the full power and capabilities of a large model like GPT, Claude, or Gemini. Just like using a sledgehammer to swat flies, using a standard LLM for simple yet high-volume tasks is excessive and wastes both money and computing resources. Small LLMs like Google’s Gemini 1.5 Flash, Meta’s Llama 3 8b, and Anthropic’s Claude 3 Haiku are able to perform these simple tasks more quickly and efficiently.

OpenAI stated that GPT-4o mini will have the same size context window of 128,000 tokens (roughly the content of a book) as the full-size version, with the same knowledge cutoff as of October 2023, although the exact size of the new model wasn’t specified. Currently, the model API offers text and vision capabilities, but video and audio will be added in the future. This announcement comes just a few weeks after OpenAI provided an update on its anticipated advanced Voice Mode as part of GPT-4o. The company mentioned that a smaller alpha release was still to come in late July, with a wider rollout planned for this fall.