For the first time in years, OpenAI is releasing new large language models that are free for developers to use and modify. It’s a major milestone for a company that has faced growing criticism for moving away from its original “open” mission. After a long delay for extra safety testing, two new “open-weight” models are now available for download.
To be clear, OpenAI isn’t releasing the full recipe for its AI. Instead of “open-source,” which would include all the training data and code, these are “open-weight” models. This means the company is sharing the final, trained “brain” of the AI, but not the secret sauce used to create it. This allows developers to use the models and even fine-tune them for their own specific needs, without giving rivals the ability to reverse-engineer OpenAI’s core technology.
The company is releasing two models, a large 117-billion parameter version and a smaller 21-billion parameter one. The smaller model is particularly interesting because OpenAI says it can run on any modern computer with 16GB of RAM, meaning you could use it to do things like write code on your machine without an internet connection.
OpenAI says these new free models are nearly as powerful as its commercial ones in some areas, like coding. The one big limitation is that they can’t handle images, video, or voice. For that, you’ll still need to use OpenAI’s paid cloud services.
This move is a big deal in the AI world. It “democratizes” access to powerful AI, letting more people build on top of it without needing massive, expensive data centers. The release is also somewhat embarrassing for Meta, which has recently signaled it will release fewer open models, a major reversal of its previous stance.
