Researchers who need to trace model behaviour back to training data get unusually open weights, data and checkpoints at no cost.
Kenning's take
What it does
Olmo 3 is a family of open language models from the Allen Institute for AI (Ai2) in 7 billion and 32 billion parameter sizes, including Base, Think (reasoning), Instruct (chat) and RL Zero versions. Ai2 publishes the weights together with the training data, code and intermediate checkpoints, so researchers can trace behaviour back to the data. People can try it in the Ai2 Playground, download it from Hugging Face or use it through OpenRouter; an Olmo 3.1 update followed in December 2025.
- 7B and 32B sizes
- Think models that show their reasoning steps
- Instruct model for chat and tool use
- Training data, code and checkpoints all released
- Context of up to about 65K tokens
- Olmo 3.1 Think and Instruct 32B update
Who it suits. Researchers and developers who want an open model whose full training process can be inspected and changed.
Key facts
- Category
- AI models
- Made by
- Allen Institute for AI (Ai2)
- Released
- 20 November 2025
- How to get it
- free in the Ai2 Playground, as downloads on Hugging Face, and through OpenRouter and other inference partners
- Works on
- Web (Ai2 Playground), API, Self-hosted
- In the UK
- Yes
- Free plan
- free weights and free Ai2 Playground
FAQ
When was Olmo 3 released?
Ai2 released Olmo 3 on 20 November 2025, and added Olmo 3.1 checkpoints in December 2025.
Is Olmo 3 free?
Yes. The weights, training data and code are free to download, and the models can be tried in the Ai2 Playground.
What sizes does Olmo 3 come in?
7 billion and 32 billion parameters.
Facts from the maker's own pages and the sources listed. Checked 2026-10-11.
Similar tools
ACE-Step 1.5
Open-source music generation model that writes full songs with lyrics and runs on ordinary consumer GPUs.
AlphaGenome Atlas
Google DeepMind database of AI-predicted effects for every possible single-letter change in human DNA.
Amazon Nova 2
Amazon's second family of Nova AI models, covering reasoning, speech conversation and combined text-and-image generation.
Cisco Antares
Cisco's small open-weight models that find where known security vulnerabilities sit in a codebase, running locally.