In November 2025, the Allen Institute for AI released Olmo 3, a family of 7B and 32B language models. What sets it apart isn't raw benchmark scores but honesty about how it was built. Most "open" models hand you the finished weights and nothing else. Olmo 3 ships the entire model flow: the training corpus, the training code, intermediate checkpoints, and an Apache 2.0 license that lets you actually use it commercially.
A companion tool lets you connect a model's output back to the specific training data behind it. That kind of visibility is rare, and it matters for anyone who needs to explain why an AI system said what it said.
For a small or mid-size business, the takeaway isn't that you should self-host a language model tomorrow. It's that fully open models give you a real fallback. If a vendor changes pricing or terms, an Apache-licensed model you can run yourself is a real fallback. Worth knowing it exists before you need it.
Further reading: original source ↗
Want this handled for you?
This is exactly the kind of thing our AI integration & automation work takes off your plate. Start with a free, no-pressure look at your site.
Come aboard