Overview
PP-OCRv6_small_det is the small text-detection model in PaddleOCR’s PP-OCRv6 series. It locates text in images; it is not a standalone text-transcription or generative language model. The model has approximately 2.48 million parameters and uses a PPLCNetV4 backbone with a RepLKFPN neck. It is intended to balance accuracy and efficiency for mobile deployment. (huggingface.co)
Capabilities and uses
- Detects multilingual text regions, including handwritten, printed, rotated, curved and artistic text. Potential uses include OCR for documents, photographs and industrial images. (huggingface.co)
- Can be combined with a separate recognition model, such as PP-OCRv6_small_rec, in a PaddleOCR pipeline to transcribe detected text. The PP-OCRv6 small-tier OCR solution supports 50 languages; that figure describes the OCR solution, not transcription by the detector alone. (huggingface.co)
- The official detection-module documentation reports 84.1% detection Hmean and a 9.6 MB model size. These are published evaluation results, not guarantees for every dataset or deployment. (paddleocr.ai)
Limitations and deployment
The detector alone does not output recognized character strings; a recognition stage is required. Detection can miss text or produce inaccurate boxes, depending on the image and settings. The official ONNX model card lists the weights under Apache-2.0 and describes local use through PaddleOCR and ONNX Runtime. (huggingface.co)
Release date
June 11, 2026: PaddleOCR 3.7.0 announced the PP-OCRv6 series. This is the official series release date; a separate release date for this exact detector was not established. (github.com)
Pricing
- Standalone detector: No separate official input or output token price was verified. The downloadable model can be self-hosted; infrastructure and operational costs are separate.
- Related mainland China service, CNY: Baidu’s PaddleOCR Enterprise PP-OCRv6 OCR service lists postpaid usage at CNY 0.025 per successful call, with no monthly volume tier; failed calls are not billed. This is a complete OCR service price, not a verified price for PP-OCRv6_small_det alone. It is billed per call rather than separately for input and output tokens. (ai.baidu.com)
- Prepaid service packages, CNY: 10,000 calls for CNY 240; 50,000 for CNY 1,100; 100,000 for CNY 2,000; 200,000 for CNY 3,400; 500,000 for CNY 6,800; 1,000,000 for CNY 11,000; and 5,000,000 for CNY 45,000. Packages expire one year after purchase. Verified individual and enterprise users receive 1,000 and 2,000 trial calls respectively. These terms apply to the related PP-OCRv6 service, not a confirmed detector-only offering. (ai.baidu.com)
- International official USD pricing: Not verified for this detector. No currency conversion or third-party resale price has been used.
Official mainland pricing source: Baidu AI Open Platform pricing documentation. Pricing checked November 16, 2026.