MolmoWeb 8B

By our estimate, MolmoWeb 8B needs about 6.5 GB at Q4, 11.1 GB at Q8 and 20.4 GB at FP16 of VRAM. That covers the weights plus headroom for a short context, and longer contexts need more.

Yes, by our estimates MolmoWeb 8B fits on an 8 GB GPU at Q4 (about 6.5 GB) and at higher precision on larger cards.

Yes, a Mac with at least 16 GB of unified memory should run MolmoWeb 8B at Q4, which needs an estimated 6.5 GB. This assumes macOS gives the GPU about 75% of unified memory.

Yes, MolmoWeb 8B has open weights, so you can run it on your own hardware and keep prompts on your own machines.

Run MolmoWeb 8B locally

Yes, MolmoWeb 8B accepts images as input alongside text.

Similar models from other labs include Nemotron Cascade 2 30B A3B (NVIDIA), MiniMax-M2.7-highspeed (MiniMax) and Nemotron 3 Content Safety (NVIDIA). They are the closest matches by kind and release date.

Nemotron Cascade 2 30B A3BMolmoWeb 8B vs Nemotron Cascade 2 30B A3BMiniMax-M2.7-highspeedMolmoWeb 8B vs MiniMax-M2.7-highspeedNemotron 3 Content SafetyMolmoWeb 8B vs Nemotron 3 Content Safety

MolmoWeb 8B has open weights under the Apache 2.0 license, so you can download and run it yourself.

Yes, MolmoWeb 8B is released under the Apache 2.0 license, which allows commercial use. Check the license text for attribution terms.

The MolmoWeb 8B weights are on Hugging Face at allenai/MolmoWeb-8B.

Run MolmoWeb 8B locally

MolmoWeb 8B was made by Ai2. It was released on March 24, 2026.

More from Ai2

Yes, MolmoWeb 8B has open weights, so you can run it on your own hardware without API fees.