Ai2 (Allen Institute for AI)MolmoPoint
MolmoPoint 8B
MolmoPoint 8B grounds objects in images, image sets and video.
What the release provides
It emits special pointing tokens instead of plain coordinate text. Decoding needs metadata from the preprocessor and the model's point-extraction functions. The Hugging Face implementation supports inference; the separate repository provides training code.MolmoPoint 8B official model card
How to use it well
Practical guidance. Preserve preprocessing metadata with each request. A decoded point must be mapped back to the correct source image, frame and object identifier before it is used.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | allenai/MolmoPoint-8B |