OpenAI today announced a lower-cost “mini” model that it says will allow more companies and programs to benefit from AI. The fresh model, called GPT-4o mini and available today, is 60 percent cheaper than OpenAI’s cheapest existing model while offering higher performance, the company says.
OpenAI is characterizing the move as part of an effort to make AI “as widely accessible as possible,” but it also reflects growing competition among AI cloud providers, as well as growing interest in tiny and free open-source AI models. Meta is expected to debut the largest version of its high-performance free offering, Llama 3, next week.
“The whole point of OpenAI is to build and distribute AI safely and make it widely available,” Olivier Godement, OpenAI’s product manager behind the fresh model, tells WIRED. “Making intelligence available at lower cost is one of the most effective ways to do that.”
Godement says the company created a lower-cost offering by improving the model’s architecture and refining the training data and training scheme. The GPT-4o mini outperforms other “small” models on the market in several popular benchmarks, OpenAI says.
OpenAI has gained a significant foothold in the cloud AI market thanks to the extraordinary capabilities of its chatbot, ChatGPT, which debuted in behind schedule 2022. The company lets outsiders access the immense language model that powers ChatGPT, called GPT-4o, for a fee. It also offers a less powerful model, called GPT-3.5 Turbo, for about a tenth of the price of GPT-4o.
The interest in language models sparked by ChatGPT’s huge success has prompted competitors to develop similar offerings. AI pioneer Google has made a major effort to build and commercialize a immense language model and chatbot under the Gemini brand. Startups like Anthropic, Cohere, and AI21 have raised millions to develop and sell their own immense language models to enterprise customers and developers.
Building the most effective immense language models requires huge financial resources, but some companies have decided to open source their work to attract developers to their ecosystems. The most notable open-source AI model is Meta’s Llama; it is free to download and apply, but its license imposes some restrictions on commercial apply.
In April of this year, Meta announced Llama 3, its most powerful free model yet. The company released a tiny version of the model with 8 billion parameters—a gritty measure of the model’s portability and complexity—as well as a more effective, medium-sized version with 70 billion parameters. The medium-sized model comes close to OpenAI’s best offering in several benchmark results.
Multiple sources have confirmed to WIRED that Meta plans to release the largest version of Llama 3, with 400 billion parameters, on July 23, although they say the release date is subject to change. It’s unclear how powerful this version of Llama 3 will be, but some companies have turned their attention to open-source AI models because they’re cheaper and more customizable, and they offer more control over the model and the data fed to it.
Godement acknowledges that customer needs are evolving. “We’re seeing more and more in the market that developers and companies are combining small and large models to create the best product experience at a price and with a latency that makes sense for them,” he says.
Godement says OpenAI’s cloud offering gives customers models that have passed more security checks than competitors. He adds that OpenAI could eventually develop models that customers can run on their own devices. “If we see a huge demand, we could open that door,” he says.