InferX Beta Serverless GPU Inference Platform, Built for Agent-Native Workloads

DeepSeek-OCR

DeepSeek-OCR (released in late 2025, with v2 arriving in January 2026) is a specialized multimodal model designed to solve the "token explosion" problem in traditional Document AI. While standard Vision-Language Models (VLMs) often convert a single page into thousands of tokens, DeepSeek-OCR treats OCR as a multimodal compression task, achieving high accuracy with a fraction of the computational cost.
deepseek-ai multimodal image2text
Log in to deploy: this public page shows the catalog model details, but deployment and customization stay behind login.
Log in to deploy

Metadata

Provider
deepseek-ai
Modality
multimodal
API type
image2text
Source
huggingface / deepseek-ai/DeepSeek-OCR
Created
2026-04-07 00:16:38 UTC
Updated
2026-04-07 00:16:38 UTC
Catalog version
1
Visibility
Published

Specifications

Parameters
3.00B
MoE
No
Max model length
2000
Image
vllm/vllm-openai:v0.13.0

Default Deploy Config

GPU count
1
vRAM
20000 MB
Summary
1xGPU 20000 MB

Recommended Use Cases

Model Spec