MolmoAct 7B-D

By our estimate, MolmoAct 7B-D needs about 6.2 GB at Q4, 10.4 GB at Q8 and 19.2 GB at FP16 of VRAM. That covers the weights plus headroom for a short context, and longer contexts need more.

Yes, by our estimates MolmoAct 7B-D fits on an 8 GB GPU at Q4 (about 6.2 GB) and at higher precision on larger cards.

Yes, a Mac with at least 16 GB of unified memory should run MolmoAct 7B-D at Q4, which needs an estimated 6.2 GB. This assumes macOS gives the GPU about 75% of unified memory.

Yes, MolmoAct 7B-D has open weights, so you can run it on your own hardware and keep prompts on your own machines.

Run MolmoAct 7B-D locally

Yes, MolmoAct 7B-D accepts images as input alongside text.

Similar models from other labs include Baichuan M2 32B (Baichuan), Nemotron Nano 9B v2 (NVIDIA) and Seed-OSS-36B-Instruct (ByteDance). They are the closest matches by kind and release date.

Baichuan M2 32BMolmoAct 7B-D vs Baichuan M2 32BNemotron Nano 9B v2MolmoAct 7B-D vs Nemotron Nano 9B v2Seed-OSS-36B-InstructMolmoAct 7B-D vs Seed-OSS-36B-Instruct

MolmoAct 7B-D has open weights under the Apache 2.0 license, so you can download and run it yourself.

Yes, MolmoAct 7B-D is released under the Apache 2.0 license, which allows commercial use. Check the license text for attribution terms.

The MolmoAct 7B-D weights are on Hugging Face at allenai/MolmoAct-7B-D-0812.

Run MolmoAct 7B-D locally

MolmoAct 7B-D was made by Ai2. It was released on August 12, 2025.

More from Ai2

Yes, MolmoAct 7B-D has open weights, so you can run it on your own hardware without API fees.