What just happened? OpenAI is ready to deprecate GPT 3.5, the AI model it released to the public in late 2022 alongside the popular ChatGPT service. The LLM will be replaced by GPT-4o mini, a smaller model that the company claims outperforms industry-leading compact offerings from the same category in various reasoning tasks involving text and visuals.
The key advantages of GPT-4o mini are its cost-effectiveness and greater speed compared to OpenAI’s cutting-edge AI models, such as GPT-4 Omni, launched in May. The latest mini model has a context window of 128,000 tokens, roughly the length of a book, and a knowledge cutoff of October 2023.
According to OpenAI, GPT-4o mini surpasses its predecessor, GPT-3.5 Turbo, and other small models on academic benchmarks that test textual intelligence and multimodal reasoning. The new mini model also shines when it comes to function calling, which enables developers to build applications that can fetch data or take actions with external systems. It supports the same range of languages as GPT-4o and boasts improved long-context performance compared to its older sibling.
When compared to the competition, OpenAI claims that GPT-4o mini scores 82% on MMLU, a benchmark that measures reasoning, compared to 79% for Gemini 1.5 Flash and 75% for Claude 3 Haiku.
On MGSM, which tests math reasoning, GPT-4o mini scored a solid 87%, while Flash and Haiku trailed behind with 78% and 72%, respectively.
The biggest perk is the processing cost. OpenAI says GPT-4o mini is significantly more affordable to run than its previous models and more than 60% cheaper than GPT-3.5 Turbo. For developers, that translates to a cost of just 15 cents per million input tokens and 60 cents per million output tokens.
Compact AI models are becoming an increasingly popular choice for developers seeking efficient solutions for high-volume, simple tasks that require repeated AI model interactions. Better yet, they consume significantly less power, making them more environmentally friendly.
GPT-4o mini is now available to developers through OpenAI’s API, as well as to consumers via the ChatGPT web and mobile app. It’s set to become the new default model for free tier users of ChatGPT. Enterprise users will get their hands on it next week, the company says.
Looking ahead, the company envisions a future where AI models are seamlessly integrated into every app and website, and they’re committed to driving down costs while enhancing model capabilities.
