Positioning
PP-OCRv6_tiny_rec is the lightweight text-recognition component of PaddleOCR’s PP-OCRv6 family, intended for resource-constrained edge and IoT deployment. It uses a PPLCNetV4 backbone with LightSVTR and a CTC/NRTR multi-head decoder. The documented model storage size is 4.4 MB. (paddleocr.ai)
Capabilities and uses
The tiny tier supports 49 languages, including Simplified Chinese, Traditional Chinese, English and Latin-script languages. It can be used to transcribe text regions in images as part of an OCR workflow, including document digitization and industrial text recognition. On PaddleOCR’s internal multi-scenario recognition benchmark, the tiny model achieved 73.5% weighted average accuracy; actual results depend on the input and deployment conditions. (paddleocr.ai)
Limitations
The tiny tier does not officially support Japanese. It is a recognition component rather than a complete text-detection and document-parsing pipeline; full-page OCR generally requires a separate detection stage. Its reported recognition accuracy is lower than the small and medium tiers on PaddleOCR’s internal benchmark. (paddleocr.ai)
Release date
June 11, 2026, as part of the PP-OCRv6 release and PaddleOCR 3.7.0 announcement. (github.com)
Pricing
No separate official input-token or output-token price for PP-OCRv6_tiny_rec was verified. The downloadable model is intended for local deployment; infrastructure and operating costs depend on the deployment.
For context only, Baidu AI Cloud’s mainland China PP-OCRv6 enterprise API lists pay-as-you-go pricing of CNY 0.025 per successful call, with no separate input/output token charge stated. This is a price for the hosted PP-OCRv6 service, not a verified price for the tiny recognition model; the service documentation does not establish that it uses this specific variant. Pricing checked November 16, 2026. (bce-cdn.bj.bcebos.com)