The fully open language model — weights, data, and code, all published.
AI2's OLMo is the rare model released with everything: weights, training data, code, and logs. OLMo 2 reaches 32B-class quality while letting anyone reproduce or extend the work from scratch.
Who it's for: Researchers, academics, and orgs that require fully reproducible, auditable, self-hosted models.
OLMo publishes training data, code, and logs — not just weights — for true reproducibility.
The second generation rivals much larger closed models on standard benchmarks.
Built so researchers can study, ablate, and extend every training decision.
Run it on your own hardware with standard open tooling.
OLMo is the gold standard for teams that must audit and reproduce every part of a model. If you just want an API, a hosted model is easier — but for science and sovereignty, nothing is more open.