Model
Model explorer

QVQ-72B-Preview

OPEN
Alibaba · QVQ family · released Dec 25, 2024

Experimental step-by-step visual reasoning model on a Qwen2-VL-72B base; single-round dialogue, no video input.

ReasoningCodingVisionFunction callingTool useAgentic
1127.6
Elo · rank #185
Parameters
73B
Active params
73B (dense)
Context
32K tokens
Architecture
Dense Vision-Language Transformer (built on Qwen2-VL-72B)
License
Qwen (Tongyi Qianwen) License
Languages
—
API price (in/out)
No hosted API
Modalities
text · vision
Benchmark results
Bar shows position within the tracked field; marker = field best
ChartQAVision83.2%#26
best: MiniMax-VL-01 · 91.7%
DocVQAVision92.7%#26
best: Qwen2-VL-72B · 96.5%
Flickr30kVision72.96#3
best: CogVLM-17B · 94.9
GQAVision63.9%#1
best: this model · 63.9%
MathVisionVision34.9%#17
best: Seed 2.1 Pro · 92.6%
MathVistaVision70.3%#27
best: Seed 2.1 Pro · 90.7%
MMMUVision70.3%#52
best: Claude Fable 5 · 89.3%
OlympiadBenchMath20.4%#5
best: Doubao-1.5-Pro · 59.8%
Run it locally
VRAM @ Q4
47 GB
VRAM @ FP16
146 GB
Fits on (Q4)
M3 Max 128GBM3 Ultra 512GBA100 80GBH100 80GBH200 141GBB200 192GB
Throughput data unavailable.
Quantizations
GGUF
QVQ family
Elo progression across releases
QVQ-72B-PreviewDec 20241127.6
Resources
HFQwen/QVQ-72B-PreviewDOCSqwenlm.github.io/blog/qvq-72b-previewAPIAvailable via Alibaba API
API price weights · each benchmark row carries its own source badge (see methodology)