OpenAI is introducing GPT-4o Mini, a lighter and more affordable model aimed at making AI accessible to a broader range of developers. This new model, which is cheaper and more capable than GPT-3.5, represents OpenAI’s effort to democratize AI development. Building apps with OpenAI’s models can be expensive, often pricing out smaller developers who then turn to more affordable alternatives like Google’s Gemini 1.5 Flash or Anthropic’s Claude 3 Haiku. Now, with GPT-4o Mini, OpenAI is entering the light model game, offering a cost-effective solution for developers.
“I think GPT-4o Mini really gets at the OpenAI mission of making AI more broadly accessible to people. If we want AI to benefit every corner of the world, every industry, every application, we have to make AI much more affordable,” Olivier Godement, who leads the API platform product, told The Verge.
Starting today, ChatGPT users on Free, Plus, and Team plans can use GPT-4o Mini instead of GPT-3.5 Turbo, with Enterprise users gaining access next week. This means GPT-3.5 will no longer be an option for ChatGPT users but will still be available via the API for developers who prefer not to switch to GPT-4o Mini. However, Godement indicated that GPT-3.5 will eventually be retired from the API.
The new, lightweight model supports both text and vision in the API and is expected to handle all multimodal inputs and outputs, including video and audio, in the near future. This could lead to more capable virtual assistants that can understand and manage tasks like travel itineraries, though GPT-4o Mini is designed for simpler tasks.
GPT-4o Mini achieved an 82 percent score on the Measuring Massive Multitask Language Understanding (MMLU) benchmark exam, which consists of about 16,000 multiple-choice questions across 57 academic subjects. This score surpasses GPT-3.5’s 70 percent but falls short of GPT-4o’s 88.7 percent and Google’s Gemini Ultra’s 90 percent. Competing models Claude 3 Haiku and Gemini 1.5 Flash scored 75.2 percent and 78.9 percent, respectively.
It’s important to note that benchmark tests like the MMLU have their limitations. Variations in how different companies administer these tests can make direct comparisons challenging, as highlighted by The New York Times. Additionally, the possibility of AI models having pre-existing knowledge of the test questions can skew results, and typically, no third-party evaluators are involved in the process.
For developers eager to build AI applications cost-effectively, GPT-4o Mini offers a valuable new tool. OpenAI allowed the financial technology startup Ramp to test the model, using GPT-4o Mini to create a tool that extracts expense data from receipts. Instead of manually inputting data, users can upload a picture of their receipt, and the model sorts it automatically. Similarly, the email client Superhuman tested GPT-4o Mini to develop an auto-suggestion feature for email responses.
The goal of GPT-4o Mini is to provide a lightweight and inexpensive option for developers to create applications and tools that were previously unaffordable with larger models like GPT-4. Many developers previously turned to Claude 3 Haiku or Gemini 1.5 Flash due to the high costs associated with more robust models.
When asked why OpenAI took so long to release a lighter model, Godement attributed it to “pure prioritization,” as the company was initially focused on developing larger and more advanced models like GPT-4, which required significant resources. Over time, OpenAI noticed a growing demand for smaller models among developers, prompting the investment in GPT-4o Mini.
“I think it’s going to be very popular,” Godement said. “Both by existing apps that use all the AI at OpenAI and also many apps that were put out by the pricing before.”
Your First 10 AI Skills
10 practical AI skills, copy-paste prompts and a 7-day plan to start using AI with confidence.
Download the guide →