Model

Molmo 72B

OPEN

Allen Institute for AI (Ai2) · Molmo family · released Sep 25, 2024

Pioneered 2D image-pointing; trained on 1M curated PixMo pairs with 1000x less data than peers, rivals GPT-4o/Claude 3.5/Gemini 1.5

ReasoningCodingVisionFunction callingTool useAgentic

1106.0

Elo · unrated

Parameters

72B

Active params

72B (dense)

Context

4K tokens

Architecture

Dense VLM: Qwen2-72B LLM backbone + OpenAI CLIP ViT-L/14-336 vision encoder

License

Apache-2.0

Languages

—

API price (in/out)

No hosted API

Modalities

text · vision

Benchmark results

Bar shows position within the tracked field; marker = field best

AI2DVision96.3%#1

best: this model · 96.3%

ChartQAVision87.3%#13

best: MiniMax-VL-01 · 91.7%

DocVQAVision93.5%#16

best: Qwen2-VL-72B · 96.5%

MathVistaVision58.6%#44

best: Seed 2.1 Pro · 90.7%

MMMUVision54.1%#87

best: Claude Fable 5 · 89.3%

TextVQAVision83.1%#8

best: Molmo 2 8B · 85.7%

VQAv2Vision86.5%#3

best: Reka Edge (2026) · 88.4%

Run it locally

VRAM @ Q4

39 GB

VRAM @ FP16

144 GB

Fits on (Q4)

M3 Max 128GBM3 Ultra 512GBA100 80GBH100 80GBH200 141GBB200 192GB

72B model exceeds a single RTX 4090's 24GB VRAM even at 4-bit (~39GB); needs 48GB+ or multi-GPU

Quantizations

bitsandbytes NF4

Molmo family

Elo progression across releases

Molmo 72BSep 20241106.0

Resources

HFallenai/Molmo-72B-0924 GHallenai/molmo DOCSmolmo.allenai.org

API price weights · each benchmark row carries its own source badge (see methodology)