Molmo
Family overview — summarizes releases; licenses and availability belong to each release.
Maintained by Ai2 (Allen Institute for AI)126
Molmo is Ai2's family of open vision-language models. The first release (September 2024) comprised MolmoE-1B, Molmo-7B-O, Molmo-7B-D, and Molmo-72B, trained with Ai2's PixMo data. Molmo 2 (December 2025) added video and multi-image understanding and grounding in 4B and 8B models built on Qwen3 and a 7B model (Molmo2-O-7B) built on Ai2's own Olmo 3. Ai2 has since built specialized variants on the line, including MolmoAct 2 for robot manipulation (May 2026).1235
- Release notes: Molmo 2 announcement (external site: allenai.org)
- Paper: Molmo 2 paper (arXiv 2601.10611) (external site: arxiv.org)
- Repository: Molmo 2 code (allenai/molmo2) (external site: github.com)
- Release notes: Original Molmo announcement (external site: allenai.org)
Releases assessed
| Release | Weights | Tier (USASI rubric v0.1) | License | Released |
|---|---|---|---|---|
| Molmo2-O 7B | Weights: Public | Model-disclosure tier (USASI rubric v0.1): Open-stack | Apache-2.0, Apache-2.0 | Dec 2025 |
What it is useful for
Organization context
Provenance and derivatives
Molmo models pair a pretrained vision encoder with a pretrained language model and are trained on Ai2's own image and video datasets (PixMo for the first release, the Molmo2 datasets for Molmo 2). Molmo2-4B and Molmo2-8B use the Qwen team's Qwen3 language models; Molmo2-O-7B uses Ai2's Olmo 3 7B Instruct; the Molmo 2 cards name SigLIP 2 (published under Google's Hugging Face organization) as the vision backbone.67814
- Derived from: Olmo 3 7B Instruct — Language model for Molmo2-O-7B.
- Derived from: Qwen3-8B (external site: huggingface.co) — Language model for Molmo2-8B, from the Qwen team; not developed by Ai2.
U.S. eligibility
Eligible · basis: U.S. nonprofit or lab
The Molmo models are developed by Ai2 (Allen Institute for AI), which describes itself as a Seattle-based non-profit AI research institute. Eligibility of each release is assessed on the release record; some releases build on third-party language models, recorded as provenance.69
Sources
This listing is not an endorsement, a safety assessment, or a federal approval.