vram.run Models Hardware Providers Cloud State of Inference
API provider data is live · Hardware & cloud pricing curated 2026-02-23

DeepSeek-OCR

by deepseek-ai

3.3B params · image-text-to-text · 3.3k likes · 2.3M downloads

DeepSeek-OCR is a 3.3B parameter model. At Q4 quantization it requires 2GB of VRAM. It runs comfortably on GeForce RTX 4090 (393 tok/s), GeForce RTX 5090 (590 tok/s), M4 Max 128GB (144 tok/s).

Inference providers

Provider$/1M in$/1M outThroughput
Novita203 tok/s

GPU compatibility

GPUVRAMQ4 DecodeVerdict
GeForce RTX 409024GB393 tok/scomfortable
GeForce RTX 509032GB590 tok/scomfortable
M4 Max 128GB128GB144 tok/scomfortable
M4 Pro 48GB48GB72 tok/scomfortable
M4 Pro 24GB24GB72 tok/scomfortable
A100 PCIe 80 GB80GB721 tok/scomfortable
H100 SXM5 80 GB80GB1450 tok/scomfortable
GeForce RTX 309024GB347 tok/scomfortable
Radeon RX 7900 XTX24GB287 tok/scomfortable
GeForce RTX 408016GB279 tok/scomfortable
Install CLI [email protected] Raw data · MIT · API data: live · HW/Cloud data: curated 2026-02-23 · v0.6.0