[{"data":1,"prerenderedAt":4},["ShallowReactive",2],{"readme:turboocr":3},"\u003Cp align=\"center\">\n  \u003Csub>🧪 Apple Metal and Intel OpenVINO backends are in testing, with AMD ROCm in development. NVIDIA is the only backend shipped today.\u003C\u002Fsub>\n\u003C\u002Fp>\u003Cp align=\"center\">\n  \u003Cimg src=\"https:\u002F\u002Fraw.githubusercontent.com\u002Faiptimizer\u002Fturboocr\u002FHEAD\u002Ftests\u002Fbenchmark\u002Fcomparison\u002Fimages\u002Fbanner.png\" alt=\"TurboOCR — the fastest GPU document parser.\" width=\"100%\" \u002F>\n\u003C\u002Fp>\u003Cp align=\"center\">\n  \u003Cstrong>English\u003C\u002Fstrong> | \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Faiptimizer\u002Fturboocr\u002Fblob\u002FHEAD\u002FREADME_zh.md\" rel=\"nofollow ugc noopener\">简体中文\u003C\u002Fa>\n\u003C\u002Fp>\u003Cp align=\"center\">\n  \u003Cstrong>The fastest GPU document parser — OCR · layout · tables · formulas → Markdown, at 200–559 images\u002Fs on one GPU.\u003C\u002Fstrong>\u003Cbr \u002F>\n  C++ \u002F CUDA \u002F TensorRT \u002F PP-OCRv6 — Linux + NVIDIA GPU\n\u003C\u002Fp>\u003Ch3>🎉 v3.0 — now powered by PP-OCRv6\u003C\u002Fh3>\n\u003Cp align=\"center\">\n  \u003Csub>New \u003Ccode>medium\u003C\u002Fcode> \u002F \u003Ccode>small\u003C\u002Fcode> \u002F \u003Ccode>tiny\u003C\u002Fcode> tiers · higher accuracy · faster defaults · \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Faiptimizer\u002Fturboocr\u002Fblob\u002FHEAD\u002Fdocs\u002Fbuild\u002Fupgrading-v3.md\" rel=\"nofollow ugc noopener\">breaking changes\u003C\u002Fa>\u003C\u002Fsub>\n\u003C\u002Fp>\u003Cp align=\"center\">\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Faiptimizer\u002FTurboOCR\" rel=\"nofollow ugc noopener\">\u003Cstrong>⭐ Star TurboOCR on GitHub\u003C\u002Fstrong>\u003C\u002Fa> — it helps others (and agents) find it.\n\u003C\u002Fp>\u003Cp align=\"center\">\n  \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fthroughput-up_to_559_img%2Fs-blue?style=flat-square&logo=speedtest&logoColor=white\" alt=\"up to 559 img\u002Fs\" \u002F>\n  \u003Ca href=\"https:\u002F\u002Fturboocr.com\" rel=\"nofollow ugc noopener\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fwebsite-turboocr.com-3B82F6?style=flat-square&logo=googlechrome&logoColor=white\" alt=\"turboocr.com\" \u002F>\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Faiptimizer\u002FTurboOCR\u002Freleases\u002Flatest\" rel=\"nofollow ugc noopener\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fv\u002Frelease\u002Faiptimizer\u002FTurboOCR?style=flat-square&logo=github&logoColor=white\" alt=\"Release\" \u002F>\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fghcr.io\u002Faiptimizer\u002Fturboocr\" rel=\"nofollow ugc noopener\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fdocker-ghcr.io-2496ED?style=flat-square&logo=docker&logoColor=white\" alt=\"Docker\" \u002F>\u003C\u002Fa>\n  \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FC%2B%2B20-00599C?style=flat-square&logo=cplusplus&logoColor=white\" alt=\"C++20\" \u002F>\n  \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FCUDA-76B900?style=flat-square&logo=nvidia&logoColor=white\" alt=\"CUDA\" \u002F>\n  \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FTensorRT-10.16-76B900?style=flat-square&logo=nvidia&logoColor=white\" alt=\"TensorRT 10.16\" \u002F>\n  \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FgRPC-4285F4?style=flat-square&logo=google&logoColor=white\" alt=\"gRPC\" \u002F>\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\" rel=\"nofollow ugc noopener\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPP--OCRv6-PaddleOCR-0053D6?style=flat-square&logo=paddlepaddle&logoColor=white\" alt=\"PaddleOCR\" \u002F>\u003C\u002Fa>\n  \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Flicense-MIT-blue?style=flat-square&logo=opensourceinitiative&logoColor=white\" alt=\"MIT License\" \u002F>\n\u003C\u002Fp>\u003Cp align=\"center\">\n  \u003Ca href=\"#quick-start\" rel=\"nofollow ugc noopener\">Quick Start\u003C\u002Fa> ·\n  \u003Ca href=\"#getting-higher-accuracy\" rel=\"nofollow ugc noopener\">Accuracy\u003C\u002Fa> ·\n  \u003Ca href=\"#benchmarks\" rel=\"nofollow ugc noopener\">Benchmarks\u003C\u002Fa> ·\n  \u003Ca href=\"#models\" rel=\"nofollow ugc noopener\">Models\u003C\u002Fa> ·\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Faiptimizer\u002Fturboocr\u002Fblob\u002FHEAD\u002Fdocs\u002Fbuild\u002Fupgrading-v3.md\" rel=\"nofollow ugc noopener\">v3 changes\u003C\u002Fa> ·\n  \u003Ca href=\"#api\" rel=\"nofollow ugc noopener\">API\u003C\u002Fa> ·\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Faiptimizer\u002Fturboocr\u002Fblob\u002FHEAD\u002Fdocs\u002Findex.md\" rel=\"nofollow ugc noopener\">Docs\u003C\u002Fa>\n\u003C\u002Fp>\u003Chr \u002F>\n\u003Cp>An extremely fast GPU \u003Cstrong>document parser\u003C\u002Fstrong> — not just OCR. PP-OCRv6 detection +\nrecognition, plus layout, tables (→ HTML), formulas (→ LaTeX) and reading-order\n\u003Cstrong>Markdown\u003C\u002Fstrong>, the whole pipeline on a single multi-stream CUDA\u002FTensorRT engine,\nlocally (no VLM), behind HTTP and gRPC. Whole-page OCR runs at \u003Cstrong>up to 559 images\u002Fs\non receipts\u003C\u002Fstrong> (one RTX 5090), and full structured parsing (layout + tables + formulas)\nat \u003Cstrong>~20 pages\u002Fs\u003C\u002Fstrong> — where VLM document parsers like PaddleOCR-VL run ~1 page\u002Fs. On\nforms and receipts it is accurate and 15–90× faster than classic OCR engines.\u003C\u002Fp>\n\u003Cul>\n\u003Cli>🚀 **559 img\u002Fs (receipts) · 520 (forms) · 200+ (dense docs) on one RTX 5090 — fastest by default\u003C\u002Fli>\n\u003Cli>🎯 \u003Cstrong>Accurate on forms & receipts\u003C\u002Fstrong> — competitive with PaddleOCR-VL, PaddleOCR-Python, RapidOCR, EasyOCR and Tesseract (\u003Ca href=\"#benchmarks\" rel=\"nofollow ugc noopener\">benchmarks\u003C\u002Fa>)\u003C\u002Fli>\n\u003Cli>🧠 \u003Cstrong>PP-OCRv6\u003C\u002Fstrong> — one model covers Latin + Chinese + Japanese; pick \u003Ccode>tiny\u003C\u002Fcode> (default) \u002F \u003Ccode>small\u003C\u002Fcode> \u002F \u003Ccode>medium\u003C\u002Fcode>\u003C\u002Fli>\n\u003Cli>🌐 \u003Cstrong>More scripts\u003C\u002Fstrong> — Arabic, Cyrillic, Korean, Thai, Greek via retained PP-OCRv5 recognizers\u003C\u002Fli>\n\u003Cli>📄 **PDF n\u003C\u002Fli>\n\u003C\u002Ful>\n",1785888997613]