OpenAI is launching a simpler and cheaper model for developers to use called GPT-4o Mini. It’s much cheaper than the full-sized models and is said to work better than GPT-3.5.
Creating apps with OpenAI’s models can get very expensive. Developers who can’t afford it might have to choose cheaper options like Google’s Gemini 1.5 Flash or Anthropic’s Claude 3 Haiku. Now, OpenAI is offering a budget-friendly option.
“I think GPT-4o Mini really fits with OpenAI’s goal of making AI more accessible to everyone. If we want AI to help people everywhere and in every industry, we need to make it more affordable,” said Olivier Godement, who leads the API platform product, to The Verge.
Starting today, ChatGPT users on Free, Plus, and Team plans can use GPT-4o Mini instead of GPT-3.5 Turbo. Enterprise users will get access next week. This means GPT-3.5 will no longer be an option for ChatGPT users, but developers can still use it through the API if they don’t want to switch to GPT-4o Mini. Godement mentioned that GPT-3.5 will eventually be removed from the API, though the exact timing is unknown.
GPT-4o Mini Easy-to-Use Model
The new, lightweight model will support text and vision in the API, and the company says it will soon handle all types of inputs and outputs like video and audio. With these features, it could lead to better virtual assistants that understand your travel plans and give suggestions. However, the model is designed for simple tasks, so it’s not meant for creating a cheap version of Siri.
The new model scored 82 percent on the Measuring Massive Multitask Language Understanding (MMLU) test, which has about 16,000 multiple-choice questions from 57 academic subjects.
When the MMLU test first came out in 2020, most models didn’t do well, which was the point since the previous tests had become too easy. GPT-3.5 scored 70 percent on this test, GPT-4o scored 88.7 percent, and Google says Gemini Ultra has the highest score ever at 90 percent.
In comparison, the competing models Claude 3 Haiku and Gemini 1.5 Flash scored 75.2 percent and 78.9 percent, respectively.
It’s important to mention that researchers are cautious about benchmark tests like the MMLU because the way it’s given can differ from one company to another. This makes it hard to compare scores from different models, as reported by The New York Times.
There’s also the issue of AI models possibly having the test answers in their datasets, which can give them an unfair advantage, and usually, no third-party evaluators are involved in the process.
GPT-4o Mini AI Tool for Developers
For developers looking to create AI applications affordably, the launch of GPT-4o Mini provides a new tool to use. OpenAI let the financial technology startup Ramp test the model, using GPT-4o Mini to build a tool that extracts expense data from receipts.
This means users can upload a picture of their receipt, and the model will sort the information for them, instead of having to manually enter it. Superhuman, an email client, also tested GPT-4o Mini and used it to develop an auto-suggestion feature for email responses.
The aim is to offer a lightweight and cheap option for developers to create apps and tools they couldn’t afford with larger, pricier models like GPT-4. Many developers would choose Claude 3 Haiku or Gemini 1.5 Flash over paying the high costs to run one of the most powerful models.
So, why did it take OpenAI so long? Godement said it was due to “pure prioritization” as the company focused on making bigger and better models like GPT-4, which required a lot of “people and compute efforts.” Over time, OpenAI saw that developers were keen to use smaller models, so the company decided it was time to invest in building GPT-4o Mini.
“I think it’s going to be very popular,” Godement said. “Both with existing apps that use all the AI at OpenAI and also many apps that were too expensive before.”
Conclusion
GPT-4o Mini is a cheaper and simpler tool for developers to build AI apps. It makes advanced features more affordable, helping developers create useful tools without spending too much. While it’s not a replacement for more powerful models, it’s great for simpler tasks and budget-friendly projects.
