Baidu's 0.9B PaddleOCR-VL 1.5 Just Beat GPT-4o at Reading Documents—But Who's Cashing In?
Everyone figured bloated giants like GPT-4o owned document parsing. Baidu's scrappy 0.9B model just flipped the script—94.5% accuracy, cheaper, faster. But is it hype or hardware shift?
⚡ Key Takeaways
- PaddleOCR-VL 1.5's 0.9B model hits 94.5% on OmniDocBench, topping GPT-4o with polygon layout seg and native res encoding. 𝕏
- Hybrid arch fixes traditional OCR flaws—irregular shapes, reading order—runs cheap on consumer hardware. 𝕏
- Baidu's efficiency play signals shift to sub-2B doc models, leapfrogging US giants like Tesseract 2.0. 𝕏
Worth sharing?
Get the best AI stories of the week in your inbox — no noise, no spam.
Originally reported by Towards AI
Related Stories
AI Research
Phi-4-reasoning-vision: The 15B Brain That Sees Math Problems and Crushes Big VLMs
Large Language Models
Why LLM Agents Are Secretly Bankrupting Your AI Startup
Large Language Models
Chandra OCR 2's 5B Params Smoke GPT-4o on Doc Benchmarks—Open-Source Finally Wins
Large Language Models