--- title: A 34.5M OCR Model Just Beat a 235B Giant | daily.dev description: PaddlePaddle released PP-OCRv6, a lightweight OCR model series with tiny (1.5M), small (7.7M), and medium (34.5M) parameter variants. The medium model achieves... canonical: https://daily.dev/posts/a-34-5m-ocr-model-just-beat-a-235b-giant-fanktarr7 twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: A 34.5M OCR Model Just Beat a 235B Giant | daily.dev og:description: PaddlePaddle released PP-OCRv6, a lightweight OCR model series with tiny (1.5M), small (7.7M), and medium (34.5M) parameter variants. The medium model achieves... og:url: https://daily.dev/posts/a-34-5m-ocr-model-just-beat-a-235b-giant-fanktarr7 og:image: https://api.daily.dev/og/posts/FankTaRR7.png og:image:alt: A 34.5M OCR Model Just Beat a 235B Giant og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # A 34.5M OCR Model Just Beat a 235B Giant **[bytefer](https://daily.dev/sources/fx5kjjichwjnx5totxmtp)** · [@bytefer](https://daily.dev/bytefer) · 3 min read · 1 upvotes · 0 comments ## Summary PaddlePaddle released PP-OCRv6, a lightweight OCR model series with tiny (1.5M), small (7.7M), and medium (34.5M) parameter variants. The medium model achieves 86.2% detection Hmean and 83.2% recognition accuracy, outperforming large VLMs like Qwen3-VL-235B, Gemini-3.1-Pro, and GPT-5.5 on OCR benchmarks. It supports 50 languages, handles complex scenarios (rotated text, dense text, low resolution), and runs via Paddle Inference, ONNX Runtime, and Hugging Face Transformers. A JavaScript SDK (ppu-paddle-ocr) enables local deployment on Node.js, Bun, Deno, browser, and React Native with code examples provided. ## Content The PaddlePaddle team recently released its latest OCR model series, **PP-OCRv6**, which includes three versions: tiny (1.5M parameters), small (7.7M parameters), and medium (34.5M parameters). In internal multi-scenario OCR benchmark tests, the PP-OCRv6_medium model achieved a detection Hmean of 86.2% and a recognition accuracy of 83.2%. It outperformed several large vision-language models — such as Qwen3-VL-235B, Gemini-3.1-Pro, and GPT-5.5 — on the corresponding OCR metrics.   #### PP-OCRv6 Features - Supports 50 languages, including Simplified Chinese, Traditional Chinese, English, Japanese, and 46 Latin-script languages. - Supports high-precision text detection and recognition, capable of handling scenarios involving complex backgrounds, small text, dense text, rotated text, and low-resolution text. - Lightweight and cost-effective deployment; compared with large vision-language models, PP-OCRv6 features fewer parameters, lower inference costs, and faster response speeds. - Supports multiple inference frameworks, including **Paddle Inference, ONNX Runtime, and Hugging Face Transformers**. #### PP-OCRv6 Online Demo Visit [PP-OCRv6 Online Demo ](https://batchtool.com/tools/ocr) in your browser, select a local image, choose a model from the configuration panel on the right, and click the “Recognize” button to experience the document parsing capabilities of the PP-OCRv6 model. The corresponding ONNX model will be downloaded automatically upon first use.  You can also upload images containing tables; once successfully recognized, they can be exported in various formats such as **HTML, XLSX, CSV, and Markdown**. #### Local Deployment The official PP-OCRv6 documentation provides detailed instructions on running the model using PaddleOCR and Transformer; here, I will explain how to deploy the PP-OCRv6 model locally on macOS using the **ppu-paddle-ocr** SDK. This SDK supports a wide range of platforms, including Node.js, Bun, Deno, the browser, and React Native. 1. Create a project ``` mkdir paddleocr-v6 cd paddleocr-v6 npm init -y ``` 1. Install dependencies ``` pnpm add ppu-paddle-ocr onnxruntime-web # Browser pnpm add ppu-paddle-ocr onnxruntime-node # Node or Bun ``` 1. Run the PP-OCRv6 model ``` import { readFileSync } from "node:fs"; import { PaddleOcrService } from "ppu-paddle-ocr"; const service = new PaddleOcrService({ debugging: { debug: false, verbose: true, }, }); async function main() { await service.initialize(); try { const image = readFileSync("./assets/pp-ocrv6-models.jpg"); const imageBuffer = image.buffer.slice( image.byteOffset, image.byteOffset + image.byteLength, ); const result = await service.recognize(imageBuffer); console.log(result.text); } finally { await service.destroy(); } } main().catch((error) => { console.error(error); process.exitCode = 1; }); ``` The examples above demonstrate how to perform text recognition in Node.js or Bun. If you wish to run text recognition in a browser environment, you need to use the `PaddleOcrService` exported by `ppu-paddle-ocr/web`. ``` import { PaddleOcrService } from "ppu-paddle-ocr/web"; const service = new PaddleOcrService(); await service.initialize(); const file = document.getElementById("upload").files[0]; const img = new Image(); img.src = URL.createObjectURL(file); await new Promise((r) => (img.onload = r)); const canvas = document.createElement("canvas"); canvas.width = img.width; canvas.height = img.height; canvas.getContext("2d").drawImage(img, 0, 0); const result = await service.recognize(canvas); console.log(result.text); ``` In addition to supporting the latest PP-OCRv6 model, the ppu-paddle-ocr SDK also supports earlier PaddleOCR models, such as v5 and v4. ## Similar posts on daily.dev - [PP-OCRv6 on Hugging Face: 50-Language OCR from 1.5M to 34.5M Parameters](https://daily.dev/posts/pp-ocrv6-on-hugging-face-50-language-ocr-from-1-5m-to-34-5m-parameters-b5jaxussj) · Hugging Face · 0 upvotes · 0 comments - [Baidu’s PP-OCRv5 Released on Hugging Face, Outperforming VLMs in OCR Benchmarks](https://daily.dev/posts/baidu-s-pp-ocrv5-released-on-hugging-face-outperforming-vlms-in-ocr-benchmarks-ydw1ut7js) · InfoQ · 1 upvotes · 0 comments - [OCR Showdown — Evaluating Three OCR Models](https://daily.dev/posts/ocr-showdown-evaluating-three-ocr-models-ch9qm0xde) · Medium · 0 upvotes · 0 comments --- Tags: [#machine-learning](https://daily.dev/tags/machine-learning), [#nodejs](https://daily.dev/tags/nodejs), [#vlm](https://daily.dev/tags/vlm) [View this post on daily.dev](https://daily.dev/posts/a-34-5m-ocr-model-just-beat-a-235b-giant-fanktarr7) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"DiscussionForumPosting","mainEntityOfPage":"https://daily.dev/posts/a-34-5m-ocr-model-just-beat-a-235b-giant-fanktarr7","headline":"A 34.5M OCR Model Just Beat a 235B Giant","text":"PaddlePaddle released PP-OCRv6, a lightweight OCR model series with tiny (1.5M), small (7.7M), and medium (34.5M) parameter variants. The medium model achieves 86.2% detection Hmean and 83.2% recognition accuracy, outperforming large VLMs like Qwen3-VL-235B, Gemini-3.1-Pro, and GPT-5.5 on OCR benchmarks. It supports 50 languages, handles complex scenarios (rotated text, dense text, low resolution), and runs via Paddle Inference, ONNX Runtime, and Hugging Face Transformers. A JavaScript SDK (ppu-paddle-ocr) enables local deployment on Node.js, Bun, Deno, browser, and React Native with code examples provided.","url":"https://daily.dev/posts/a-34-5m-ocr-model-just-beat-a-235b-giant-fanktarr7","datePublished":"2026-07-14T04:31:19.052Z","dateModified":"2026-09-13T21:02:21.297Z","author":{"@type":"Person","name":"bytefer","url":"https://daily.dev/bytefer","image":"https://lh3.googleusercontent.com/a-/AFdZucqlbOI8UzgORH9IT5mtLClCpRPYl-bBI9zaDVVd=s100","description":"Focus on AI Agent and AI Automation.","interactionStatistic":{"@type":"InteractionCounter","interactionType":{"@type":"EndorseAction"},"userInteractionCount":2170}},"image":"https://media.daily.dev/image/upload/s--Uvk5Y4IN--/f_auto/v1784003480/posts/FankTaRR7?_a=BAMAMicg0","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"isPartOf":{"@type":"WebPage","url":"https://daily.dev/sources/fx5kjjichwjnx5totxmtp","name":"bytefer"}} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"bytefer","item":"https://daily.dev/sources/fx5kjjichwjnx5totxmtp"},{"@type":"ListItem","position":3,"name":"A 34.5M OCR Model Just Beat a 235B Giant"}]} ```