OpenAI has introduced a new, cost-effective model for developers called GPT-4o Mini. Priced considerably lower than its larger counterparts, this model is claimed to outperform GPT-3.5. For developers, building applications with existing models can be costly, often leading them to consider more affordable alternatives like Google’s Gemini 1.5 Flash or Anthropic’s Claude 3 Haiku. With the launch of GPT-4o Mini, OpenAI aims to enhance accessibility to AI technology.
Olivier Godement, who oversees the API platform product, emphasized that GPT-4o Mini aligns with OpenAI’s mission to democratize AI. "If we want AI to benefit every corner of the world, every industry, every application, we have to make AI much more affordable," he stated.
Starting today, ChatGPT users on Free, Plus, and Team plans can access GPT-4o Mini, while Enterprise users will follow next week. Although GPT-3.5 will not be available for ChatGPT users, developers can still access it through the API until it is eventually phased out.
The lightweight model supports both text and vision in its API, with plans to accommodate various multimodal inputs and outputs, including video and audio. This could lead to more capable virtual assistants that can understand travel itineraries and offer useful suggestions, although the model is designed for simpler tasks rather than full-fledged assistants like Siri.
In performance benchmarks, GPT-4o Mini achieved an 82 percent score on the Measuring Massive Multitask Language Understanding (MMLU), which consists of about 16,000 multiple-choice questions across 57 subjects. In comparison, GPT-3.5 scored 70 percent, while GPT-4o Mini surpassed it with a score of 88.7 percent. Google’s Gemini Ultra reported the highest score of 90 percent, while Claude 3 Haiku and Gemini 1.5 Flash scored 75.2 percent and 78.9 percent, respectively. However, researchers express caution regarding benchmark tests, as the testing methods can differ across companies, complicating direct comparisons.
For developers looking to create AI applications affordably, GPT-4o Mini offers a valuable new resource. Financial technology startup Ramp tested the model and developed a tool to extract expense data from receipts, allowing users to simply upload an image instead of manually entering information. Similarly, the email client Superhuman incorporated GPT-4o Mini to enhance its auto-suggestion feature for email responses.
The introduction of GPT-4o Mini aims to provide a lightweight and budget-friendly solution for developers who previously found larger models like GPT-4 cost-prohibitive. There is a growing trend among developers favoring smaller models, prompting OpenAI to allocate resources to the development of GPT-4o Mini after focusing on more sophisticated models like GPT-4.
As Godement noted, "I think it’s going to be very popular," both among existing applications and potential new ones that previously faced budget constraints.
Ranking
