OpenAI today announced a lower-cost “mini” model that will allow more companies and programs to take advantage of its service. artificial intelligenceThe new model, called GPT-4o mini, which is available starting today, is 60% cheaper than OpenAI’s cheapest existing model while offering better performance, the company says.
Open AI The company cast the move as part of an effort to make AI “as widely available as possible,” but it also reflects growing competition among AI cloud providers and growing interest in small, free, open-source AI models. Meta plans to launch the largest version of its highly capable free service, Llama 3, next week.
“OpenAI’s biggest goal is to make AI safe to build, distribute, and widely accessible,” Olivier Godement, OpenAI’s product manager in charge of the new model, told WIRED. “Making intelligence more accessible and cheaper is one of the most efficient ways for us to do that.”
Godement says the company developed a cheaper version by improving the model architecture and refining the training data and training plan. OpenAI says the GPT-4o mini outperforms other “small” models on the market on several common benchmarks.
OpenAI has gained a huge foothold in the cloud AI market thanks to its impressive chatbot capabilities. Chat GPTis set to debut in late 2022. The company is giving outsiders access to the massive language model that powers ChatGPT, called GPT-4o, for a fee, and also offers a less powerful model called GPT-3.5 Turbo, at about one-tenth the cost of GPT-4o.
The massive success of ChatGPT has sparked interest in language models and inspired competitors to build similar products: AI pioneer Google is making a big push to build and commercialize large-scale language models and chatbots under the brand name Gemini, while startups like Anthropic, Cohere, and AI21 have raised millions of dollars to develop and sell their own large-scale language models to business customers and developers.
Building the best-performing large-scale language models requires huge capital, so some companies have chosen to open-source their work to attract developers into their ecosystem. The most famous open source AI model is Meta’s LlamaIt is free to download and use, The license imposes certain restrictions on commercial use..
This April, Meta Introducing the Rama 3The company released a small version of the model with 8 billion parameters, a rough measure of model portability and complexity, and a more powerful medium version with 70 billion parameters. The medium model is close to OpenAI’s best models in several benchmark scores.
Multiple sources confirmed to WIRED that Meta plans to release the largest version of Llama 3, with 400 billion parameters, on July 23, but the release date is subject to change. It’s unclear how powerful this version of Llama 3 will be, but several companies have said they expect it to be able to run on a 100 billion-parameter machine. Focusing on open source AI models Because they are cheaper, customizable, and offer greater control over the model and the data that goes into it.
Godement acknowledges that customer needs are evolving: “We’re seeing more and more of the marketplace where developers and companies are combining small and large models to build the best product experience at a price and latency that makes sense for them,” he says.
Godement said OpenAI’s cloud service provides customers with models that have undergone more security testing than competitors, and added that OpenAI may eventually develop models that customers can run on their own devices. “If there’s a big demand, we might open that door,” he said.