LlamaIndex 转型文档领域应用研究实验室，发布 OCRia-1 与 Extract v2.6
LlamaIndex 不只做 RAG 框架了，现在自己造 OCR 模型，4B 的 OCRia-1 加上 Extract v2.6，专啃表格手写这些硬骨头，做文档处理的朋友可以看看。
LlamaIndex 宣布转型为专注文档工作的应用研究实验室，自研覆盖精度与成本平衡点的通用 OCR 模型，通过 LlamaParse 提供服务。两周前推出的 OCRia-1 是一个 4B 参数的文档解析视觉语言模型，Extract v2.6 则定位为高精度文档提取系统，可处理复杂图表、表格、表单和手写内容。基准方面，ParseBench 覆盖 150+ 模型、月下载量超 3 万，另有 ExtractBench 到 ExtractBenchPro 的提取评测体系。
Today, @llama_index is an applied research lab building models and agents for document work. We have the best general purpose OCR models at the Pareto frontier of accuracy and cost. You can use them today within LlamaParse. ✅ This includes fast and cheap models for ingesting massive volumes of documents at scale (our ‘cost-effective’ and ‘fast’ modes, along with liteparse) ✅ This also includes extremely high-quality models that will guarantee accuracy over complex charts, tables, forms, handwriting, and more. In the last two weeks, we launched OCRia-1, our 4B VLM for doc parsing and Extract v2.6 which includes the most accurate document extraction system out there. We are also fully invested in the most comprehensive, open-source benchmarks to measure the frontier of all dimensions we care about. We have 150+ models and 30k+ monthly downloads on ParseBench. We’ve also heavily invested in extraction benchmarks, from ExtractBench to ExtractBenchPro. Our vision corresponds nicely to two massive trends we see in the market: 1️⃣ The need for specialized intelligence for various tasks 2️⃣ The massive demand for unstructured context, much of which is locked within file containers Whether you’re at a startup or F500 enterprise, if you have large piles of documents you want to digitalize and automate, come talk to us! We’d be eager to learn more. llamaindex.ai/contact Your browser does not support the video tag. 🔗 View on Twitter 💬 19 🔄 1 ❤️ 38 👀 4863 📊 26 ⚡
