Tag
LlamaIndex announced the ParseBench-2 Metric Lab, an interface for customizing how document-parsing evaluation metrics and error severities are weighted, with exportable policy JSON for use on your own documents.
LlamaIndex released Boo Bench, a 179-page benchmark of document pages designed to trick parsers and vision language models into repetition loops, with a leaderboard and paper measuring how often each system comes back 'ghost-free'.
Youtu-Parsing-Omni 是一个统一的多模态解析器,用单一编码器和 OmniSchema 将文档、图像、音频与视频编码为时空结构化记录,并通过自蒸馏后训练(OSAD)在 OmniDocBench v1.6 等基准上取得最先进成绩,5B 检查点已开源。
LlamaIndex 发布 Document Parsing Protocol (DPP) 工作草案,为文本、表格、表单、图表、化学结构和乐谱等文档解析决策提供统一规则集,包含 140 条规则并提供 Claude 技能与 ChatGPT 参考文件供试用反馈。
An open-source document-parsing benchmark called OmniParse Bench was released, running 16,288 tests across 2,937 pages in 92 languages to compare OCR solutions like DeepSeek-OCR, PaddleOCR-VL, and Surya, showing no single parser wins on every document type.
OpenBMB 发布 MiniCPM-OCR Jev(2B 参数),一款 OCR 质量评估模型,可对文本、公式和表格区域的识别结果打 0-1 分,在 JevOCRBench 上决策 AUROC 达 93.6,超过多个更大的模型,并在多种 OCR 模型输出上保持稳定泛化。
百度发布 Unlimited-OCR-2,借助 SR-SWA(稀疏参考滑动窗口注意力)在有限上下文窗口内实现超长乃至无界文档解析,并同步推出 MarathonDocBench 超长文档解析基准。
Jerry Liu 宣布发布 OCRia-1,一款面向文档解析的前沿 VLM,在准确率、成本和延迟上均达到新的 SOTA,并在 ParseBench 及多个第三方基准上取得领先成绩。
该论文提出 LTEM(LLM-Friendly Table-Enhanced Markdown),一种在 Markdown 基础上最小扩展以显式编码合并单元格与多级表头的格式,并构建了表结构复杂度分类体系(F1-F6)和受控基准 STB(1,089 张表)。实验表明 LTEM 在表格理解任务上达到与 HTML 相当的准确率,同时在 token 成本与响应稳定性上优于 HTML/LaTeX。
Vik Paruchuri announces a new word confidence model for OCR verification, allowing users to see exact word positions and confidence values, even on messy scans.
SoMark released OXR, a 1.3B open-source document parsing model that tops olmOCR-Bench, OmniDocBench and DocParsingBench while delivering 1.60× DeepSeek-OCR-2's throughput on a single RTX PRO 6000. It handles footnotes, formulas, tables and handwriting across 13 element types with coordinates, powered by RSI.
MiniCPM-OCR 1.0 (1.3B) is the first OCR-focused model in the MiniCPM-V family, achieving SOTA document parsing scores on Real5-OmniDocBench while delivering 55% faster inference via DSpark decoding and day-one support for vLLM, SGLang, llama.cpp, and Ollama.
OvisOCR3 is a compact 0.8B vision-language model that unifies document parsing, general OCR, key information extraction, figure captioning, and cross-page table reconstruction via multi-teacher on-policy distillation and mixed-task reinforcement learning, achieving state-of-the-art results on benchmarks like OmniDocBench v1.6, OCRBench v2, and PureDocBench.
Alibaba released OvisOCR2, an end-to-end document parsing model with 0.8B parameters based on Qwen3.5-0.8B, topping OmniDocBench v1.6 with a score of 96.58 — the first time an end-to-end approach has surpassed all pipeline-based approaches — and it is released under the Apache 2.0 license for commercial use.
LlamaIndex 本周论文推荐介绍了 VL-DocIR 长文档检索新论文,并展示 LlamaParse Agentic Plus 在表格跨列结构解析上的准确性,可试用该工具。
OpenDocRouter by LlamaIndex adds Mistral OCR to its unified document-parsing API, offering competitive table and reading-order performance at a p50 of 1.8s per page. The service lets developers choose among many OCR models under one API, with per-page billing and ParseBench benchmarks.
LlamaIndex promotes LlamaParse's ability to cleanly extract nested tables (e.g., from Micron's earnings deck) into markdown, highlighting features like layout awareness, multimodal parsing, multilingual support, and enterprise readiness.
Tencent 发布 Youtu-Parsing-Omni,一个 5B 的全能模态解析模型,统一处理文档、图像、图表、音频和视频输入并输出结构化 JSON,在 OmniDocBench v1.6 上取得 96.96 的最佳成绩,并开源模型权重与 vLLM 插件。
LlamaIndex 发布 OpenDocRouter,一个统一的文档转 Markdown API 平台,聚合前沿与开源 OCR 模型,并通过 ParseBench 对质量与成本进行基准测试。
LlamaIndex announces the shift from traditional one-pass OCR to agentic OCR, using multi-pass verification, layout-aware reading order, and multimodal parsing for complex documents like tables and charts, with a deeper breakdown by Logan Markewich linked in the comments.