Olmo

Olmo is a family of language models from the Allen Institute for AI in Seattle. What makes it unusual is how open it is. Most “open” models publish only the finished weights, so you can run the model but cannot see what it learned from. Olmo publishes the training data, the code and the intermediate checkpoints too, which means researchers can actually study why the model behaves the way it does.

That makes Olmo the model of choice in academic work, and it turns up regularly in studies comparing model behaviour, because it is the one case where you can trace an odd output back towards its cause.