SotaOCR
FREE OCRFiles never leave your device

FREE OCR

Completely free OCR, right in your browser

Recognize PDFs and images locally with WebGPU. Your document stays on this device: only the public model files are downloaded.

Private, client-side processingMultipage PDF support

Checking this browser

Testing WebGPU, FP16 shaders, browser memory limits, and storage before loading the model.

Drag & drop a document or image

or click to choose a file from this device

PDF, PNG, JPEG, WebP, BMP, TIFF · multipage PDFs supported

Prefer not to download a large model?

The server version starts immediately, works on any device, and is better for large batches or low-memory computers.

Try server OCR

Licenses

Models and inference engine

Paddle model weights retain their original Apache 2.0 licenses. The original SotaOCR WebGPU inference engine is covered separately.

The Attribution Link License does not relicense Paddle models or other third-party materials.