Researchers wanting a fully open vision model, data and code included, get it free under Apache 2.0.
Kenning's take
What it does
Molmo is a family of open multimodal models from Ai2, the Seattle non-profit research institute. The current Molmo 2 family, announced in December 2025, comes in 4B and 8B sizes plus a fully open 7B version built on Ai2's Olmo, and adds video understanding, object tracking and pointing to exact pixels and timestamps. The weights, training data and code are public, and the models can be tried free in the Ai2 Playground.
- Molmo 2 in 4B, 8B and fully open O-7B versions
- Video question answering, captioning and tracking
- Pointing to objects in images and video
- Open weights, training data and code
- Free demo in the Ai2 Playground
Who it suits. Researchers and developers who need an open vision model they can inspect, fine-tune or run themselves.
Key facts
- Category
- AI models
- Made by
- Ai2 (Allen Institute for AI)
- Based in
- United States
- Released
- 11 December 2025
- How to get it
- free in the Ai2 Playground, as Apache 2.0 weights on Hugging Face, and through OpenRouter
- Works on
- Web, API, self-hosted
- In the UK
- Yes
- Free plan
- free open weights and free Playground demo; a free Molmo 2 8B route on OpenRouter
FAQ
Is Molmo open source?
Yes. Ai2 releases the Molmo 2 weights under Apache 2.0, along with the training data and code.
When was Molmo 2 released?
Ai2 announced Molmo 2 on 11 December 2025.
How can I try Molmo?
In the free Ai2 Playground on the web, or by downloading the weights from Hugging Face.
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.