Last updated:
OLMo
AI2's fully open language models — weights, data, code and training recipes are all public.
Model Facts
Latest model: OLMo 3.1
- Provider
- Allen Institute for AI (AI2)
- Released
- Dec 1, 2025
- Context window
- 65K tokens
- Weights
- Open
Full Model List
| Model | Released | Status | Context | Note |
|---|---|---|---|---|
| OLMo 3.1 | Dec 1, 2025 | current | 65K tokens | upgraded Think 32B and Instruct 32B; month approximate |
| OLMo 3 | Nov 1, 2025 | legacy | 65K tokens | 7B and 32B, with Base, Instruct, Think and RL Zero variants; month approximate |
- Official model page:
- allenai.org/olmo
- Other sources:
- OLMo family (LLM Reference)
- OLMo 3 (There's An AI For That)
What it is
OLMo is the Allen Institute for AI's fully open model programme. OLMo 3 (November 2025) spans 7B and 32B sizes with Base, Instruct, Think (reasoning) and RL Zero variants and a 65,536-token context; OLMo 3.1 upgrades the 32B Think and Instruct models. Everything — the Dolma 3 pretraining corpus, intermediate checkpoints, code and recipes — is published under Apache 2.0.
Where this information comes from
Details come from AI2 coverage and trackers; release months are approximate. Release dates, context windows and benchmark scores are vendor-reported unless stated, and harnesses differ between labs, so treat the numbers as indicative.
Notable traits
- Fully open: data, code, checkpoints and recipes.
- 7B model runs on a single consumer GPU.
- A reference point for transparent model development.