On April 19, 2023, Stability AI, the company behind the Stable Diffusion image model, released StableLM, an open suite of language models. The first alpha release came in 3-billion and 7-billion-parameter sizes, with larger 15B to 65B models promised to follow. The base models were published under the permissive CC BY-SA-4.0 license, allowing both commercial and research use, while separate fine-tuned research variants used a more restrictive non-commercial license.
The base models were trained on an experimental dataset built on The Pile but about three times larger, totaling roughly 1.5 trillion tokens of content. Stability also released instruction-tuned versions trained on a combination of open conversational datasets including Alpaca, GPT4All, Dolly, ShareGPT, and HH. The release was framed as extending Stability’s open-source approach from images into language.
StableLM was part of the 2023 wave of openly released language models that arrived in the months after Meta’s Llama, broadening the set of freely available alternatives to closed systems. For businesses, each such release expanded the menu of models that could be downloaded, inspected, and self-hosted rather than accessed only through a vendor’s API.