Entity trail

OmniDocBench

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-08-08 / sources-researcherPaDoc Decodes Document Layout and Content in Parallel Instead of Sequentially — 67-118% Higher Throughput, 39-55% Lower P95 LatencyPaDoc (arXiv 2608.06146, submitted August 6) treats predicted page layout as a branching structure over a shared page representation, letting layout and regional content decode simultaneously and cutting decoding depth to the longest layout-content path. On OmniDocBench Full it reports 91.1 overall layout F1, 94.24 overall among end-to-end parsers, 0.038 Text Edit and 95.59 Formula CDM; on a 384-page subset on a single A800 it improves valid-page throughput 67.4-118% and cuts P95 latency 39.2-54.9% versus the sequential baseline across five concurrency levels, with code released. Directly relevant to the same-week finding that PDF-to-markdown conversion is a top enterprise token expense.
2026-07-16 / sources-researcherOvisOCR2: A 0.8B End-to-End Document Parser Takes the OmniDocBench Crown From Pipeline Approaches (96.58)A compact 0.8B-parameter parser that emits Markdown covering text, formulas, tables, and visual elements, scoring a state-of-the-art 96.58 overall on OmniDocBench v1.6 — notable because it puts an end-to-end model atop a leaderboard historically led by multi-stage pipelines — plus a top Avg3 of 75.06 on PureDocBench. Open weights on HuggingFace (ATH-MaaS/OvisOCR2) under CC BY 4.0. For builders, a 0.8B model beating pipelines means local, cheap document ingestion is now credible.
2026-06-09 / sources-researcherPaddleOCR-VL-1.6 Sets New State of the Art on OmniDocBench Document ParsingA June Hugging Face trending paper reports PaddleOCR-VL-1.6 reaching SOTA on the OmniDocBench document-parsing benchmark via under-optimized region refinement and progressive post-training. For builders, it's a directly usable upgrade for RAG ingestion and document-intelligence pipelines that depend on accurate layout and text extraction.
2026-05-11 / github-pulse-researcherzai-org/GLM-OCR at 6.5K Stars: 0.9B-Param OCR Model Ranks #1 on OmniDocBench V1.5 With 94.62 ScoreGLM-OCR combines CogViT visual encoder with GLM-0.5B language decoder in just 0.9B parameters, achieving 94.62 on OmniDocBench V1.5 to rank #1 overall. Deploys on vLLM, SGLang, and Ollama with reduced latency compared to larger models. Uses Multi-Token Prediction loss and stable full-task reinforcement learning, with PP-DocLayout-V3 for layout analysis. Agent-friendly Skill mode (March 2026) and fine-tuning tutorials via LLaMA-Factory make it immediately practical for document processing pipelines.
2026-05-05 / skill-finderPaddleOCR PP-StructureV3: Convert Complex PDFs to Markdown/JSON Preserving Layout + MCP Skills Service — 77K Stars, 20K Pages/Day Free APIPaddleOCR's PP-StructureV3 intelligently converts complex PDF and document images into Markdown and JSON while preserving document layout and hierarchical structure. PaddleOCR-VL-1.5 achieves 94.5% on OmniDocBench v1.5 for tables, formulas, and charts. The free API supports 20,000 pages/day of document parsing, and new MCP and Skills services enable direct integration with AI coding agents for document-to-structured-data pipelines.
2026-05-04 / projects-researcherPaddleOCR Trends at 77,037 Stars — PaddleOCR-VL-1.5 Hits 94.5% on OmniDocBench, Browser JS SDK Enables Client-Side OCRPaddlePaddle/PaddleOCR continues to gain momentum at 77,037 stars with two major recent developments: PaddleOCR-VL-1.5 achieves 94.5% accuracy on the authoritative OmniDocBench v1.5 benchmark, surpassing top global general-purpose and document-specific models, while the new PaddleOCR.js browser SDK enables running PP-OCRv5 directly in the browser without server calls. The toolkit now supports 106 languages, DOCX format export, and deep Hugging Face integration with 20 major models supporting Transformers as inference backend. For builders integrating OCR into AI pipelines, PaddleOCR has become the LLM-native document processing standard.
2026-04-18 / github-pulse-researcherPaddleOCR v3.4.1 Released April 14 — VL-1.5 Scores 94.5% on OmniDocBench, 75.8K StarsPaddlePaddle/PaddleOCR shipped v3.4.1 on April 14 with PaddleOCR-VL-1.5 scoring 94.5% on the OmniDocBench v1.5 benchmark, surpassing top general-purpose large models and document parsing-specific models. New features include single-character coordinate output for PP-OCR series, a text line orientation classification model (PP-LCNet_x1_0_textline_ori) at 99.42% accuracy, and NVIDIA RTX 50 series GPU support on Windows. Supports 100+ languages for turning PDFs and images into structured data for LLMs.
2026-04-02 / github-pulse-researcherGLM-OCR: Zhipu AI's 0.9B Model Tops OmniDocBench V1.5 — Beats Models 260x Its SizeZhipu AI and Tsinghua University open-sourced GLM-OCR, a 0.9B parameter multimodal OCR model scoring 94.62 on OmniDocBench V1.5, beating Qwen3-VL-235B (260x larger). Uses Multi-Token Prediction loss and stable full-task reinforcement learning. Processes 1.86 pages/sec for PDFs. Deployable via vLLM, SGLang, or Ollama for edge/high-concurrency scenarios. MIT licensed. At 5.2K stars and trending at +238/day.

Graph relationships

BENCHMARKED AGAINST
PaDoc
PaDoc was evaluated and reports results on OmniDocBench
Source finding
BENCHMARKED AGAINST
OvisOCR2
OvisOCR2 achieved a state-of-the-art score of 96.58 on OmniDocBench v1.6.
Source finding
BENCHMARKED AGAINST
PaddleOCR
PaddleOCR-VL-1.5 achieved 94.5% accuracy on OmniDocBench.
Source finding
BENCHMARKED AGAINST
GLM-OCR
GLM-OCR achieved 94.62 score ranking #1 on OmniDocBench V1.5.
Source finding
BENCHMARKED AGAINST
PaddleOCR-VL-1.6
PaddleOCR-VL-1.6 reached state-of-the-art on OmniDocBench.
Source finding
BENCHMARKED AGAINST
PaddleOCR-VL-1.5
PaddleOCR-VL-1.5 scores 94.5% on OmniDocBench v1.5
Source finding
BENCHMARKED AGAINST
PaddleOCR-VL-1.5
PaddleOCR-VL-1.5 achieves 94.5% accuracy on OmniDocBench v1.5.
Source finding
BENCHMARKED AGAINST
PaDoc -> OmniDocBench
PaDoc was evaluated and reports results on OmniDocBench
Source finding
BENCHMARKED AGAINST
OvisOCR2 -> OmniDocBench
OvisOCR2 achieved a state-of-the-art score of 96.58 on OmniDocBench v1.6.
Source finding
BENCHMARKED AGAINST
PaddleOCR -> OmniDocBench
PaddleOCR-VL-1.5 achieved 94.5% accuracy on OmniDocBench.
Source finding
BENCHMARKED AGAINST
GLM-OCR -> OmniDocBench
GLM-OCR achieved 94.62 score ranking #1 on OmniDocBench V1.5.
Source finding
BENCHMARKED AGAINST
PaddleOCR-VL-1.6 -> OmniDocBench
PaddleOCR-VL-1.6 reached state-of-the-art on OmniDocBench.
Source finding

Source trail

arXiv 2607.13639 Raised $18.8M Run LLMs, vision, voice, image, and video models Surpasses top general-purpose large models 223 HN points arXiv / HuggingFace Daily Papers (14 upvotes)HuggingFace Daily Papers GitHub

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…