Open language models with training data, code, recipes, and checkpoints for studying and adapting model behavior.

Open models and code for combining separately trained expert modules while contributors retain control over their data's use.

A general-purpose multimodal model family that processes and generates combinations of text, images, audio, and actions.

Instruction-following language models with open data and recipes for researchers developing and comparing post-training methods.

Open multimodal models for understanding images and video, including grounded visual reasoning across multiple images.

A vision system that converts language instructions into interpretable programs using pretrained models and image operations.