NEW DELHI: Nearly two months after, OpenAI introduced to the world its most powerful AI model yet – GPT-4o, the Sam Altman-led AI powerhouse today unveiled its lighter, cheaper model for developers. The new model called GPT-4o Mini, costs significantly less than the full-sized models and it is believed to be far more capable than GPT-3.5.
OpenAI calls the new model its most cost-efficient small model. “We expect GPT-4o mini will significantly expand the range of applications built with AI by making intelligence much more affordable. GPT-4o mini scores 82% on MMLU and currently outperforms GPT-4 on chat preferences in LMSYS leaderboard,” OpenAI stated in its official blog.
The GPT-4o mini is priced at 15 cents per million input tokens and 60 cents per million output tokens. This according to OpenAI makes it more affordable than earlier frontier models and over 60 per cent cheaper than GPT-3.5 Turbo. This is particularly beneficial for developers as building apps using existing OpenAI models comes with massive costs. So far, developers have been seen to opt for cheaper models like Anthropic’s Claude 3 Haiku or Gemini 1.5 Flash from Google. With GPT-4o Mini, OpenAI is officially entering the light AI model segment.
What does the GPT-4o Mini do?
OpenAI claims that it can do a broad range of tasks with its low cost and latency, such as applications that link multiple model calls, pass on large volumes of context to the model, or interact with customers through fast, real-time text responses via support chatbots.
GPT-4o mini supports text and vision in the API. However, OpenAI has said that support for text, image, video, and audio inputs and outputs is coming to the model in the future. The new model comes with a context window of 128K tokens, supports up to 16K output tokens, and has knowledge of up to October 2023. OpenAI claims that the new model of handling non-English text is cost-effective owing to the improved tokenizer.
When it comes to performance, GPT-4o mini has surpassed GPT-3.5 Turbo and other small models’ academic benchmarks across both multimodal reasoning and textual intelligence and supports the same range of a language as GPT-4o. The new model also showcased a strong performance in tasks like function calls and this allows developers to build applications that fetch data or take actions with external systems. It also has improved long-context performance compared to GPT-3.5 Turbo.
GPT-4o mini comes with the same safety framework as GPT-4o which has been evaluated based on both human and automated assessments. The company claims that over 70 external experts in various fields have conducted evaluations to ensure that the model comes with improved safety.
OpenAI has made the GPT-4o mini available as a text and vision model in the Assistants API, Batch API, and Chat Completions API. Developers will pay up to 15 cents per 1M input tokens and 60 cents per 1M output, which is equal to about 2500 pages in a regular book. Users of ChatGPT Plus, Free, Plus, and Team will be able to access GPT-4o Mini from today in place of GPT-3.5. Meanwhile, Enterprise users will also have access starting next week.
Agencies

Agencies