PaddleOCR-VL 1.5 tops DeepSeek-OCR in local benchmarks

While DeepSeek dominates general LLM discussions, Baidu's PaddleOCR remains the pragmatic choice for fast, local document extraction.

• –PaddleOCR-VL-1.5's 0.9B parameter size makes it significantly more efficient for edge deployment and consumer hardware than larger vision models
• –DeepSeek-OCR-1 struggles with stability on noisy or low-quality scans, sometimes looping or generating gibberish instead of text
• –The addition of irregular-shaped bounding box support in version 1.5 gives PaddleOCR a major advantage on warped or curved documents
• –Both models maintain strong multilingual support, but PaddleOCR's optimization for latency makes it better suited for production environments