Olmo 3
Model ● Covered in 7 stories + Follow
Olmo 3 is an open language model in Ai2’s Olmo/Tülu model line, released in spring 2026. Recent coverage around Olmo 3 includes work using its open stack to trace sources of benchmark behaviors (such as social reasoning) to specific training-document categories, and diagnostic analyses showing the model can rely on drug-name affixes rather than underlying drug knowledge in health-related questions. It has also been involved in post-training research where tools and datasets were used to identify and reduce safety regressions tied to preference-training examples, as well as comparative studies examining how different architectures predict token types.
Updated 12 September 2026
Specifications
No specifications recorded yet.
Latest developments
Q3 2026
- How Goodfire used Ai2’s open post-training stack to trace unwanted model behavior
- How a Georgia Tech team used the open Olmo stack to trace social reasoning
- When a model reads a drug's class from its name—not its knowledge
Q2 2026
- Which tokens does a hybrid model predict better?
- Farewell Ai2
- Building accessibility tools on a truly open foundation
Q1 2026
Relationships
Products & technology
- Derived from Dolma 3 · 1 source
- PointCheck integrated with this model · 1 source
- Dolci preference dataset derived from this model · 1 source