OCR and document intelligence
Choose, test and diagnose extraction systems by the document failures that matter.
What we currently think
No single OCR model won every tested workflow. Routing remained necessary across blank pages, tables, diagrams and reading order.
Start with your question
- Which models deserve a first test?
- Which failure can this workflow not afford?
- What evidence would change the recommendation?
Start here
OCR Benchmark Leaderboard 2026 - Best Models and Workflow Fit
OCR benchmark leaderboard for 2026 covering best OCR models, OmniDocBench scores, inference speed, VRAM, and workflow fit across Hunyuan, GLM, FireRed, DeepSeek, dots.ocr-1.5, and PaddleOCR-VL-1.5.
GLM OCR vs PaddleOCR vs DeepSeek: Workflow Guide 2026
Compare GLM-OCR, PaddleOCR-VL, DeepSeek-OCR-2, Hunyuan and Mistral by speed, grounding, blank-page handling, deployment and workflow fit.
How We Benchmark OCR Models on Scan-Heavy PDFs
A practical methodology for benchmarking OCR models on scan-heavy PDFs: corpus design, scoring, visual audit steps, failure modes, and the routing rule that emerged from a 31-PDF internal pilot.
Continue exploring
LFM2.5 Encoder vs ModernBERT - CPU, GPU and Legal Document Benchmark
We compared LFM2.5-Encoder-230M and 350M with ModernBERT-base and large on CPU speed, RTX 3090 Ti latency, memory release, sentiment classification and long legal-document classification. Here are the results and their limits.
NuExtract3 Table OCR - How We Validate Numbers and Cell Structure
A public-safe FireRed and NuExtract3 comparison showing why table OCR needs numeric-token, row, column, and cell-position checks instead of visual review alone.
Surya OCR 2 Layout Analysis - Why Higher F1 Did Not Make It Our Default
Surya OCR 2 beat Paddle on layout F1 in two public-safe tests, but weaker label coverage, fragmented boxes, and fallback behavior kept it out of our default OCR pipeline.
DeepSeek OCR-2 in Production - What the Benchmarks Don't Tell You
Production notes from running DeepSeek OCR-2 on 50 scanned pages. Covers markdown output quality, blank detection strength, and where it falls behind GLM and Qianfan. First-party CER data included.
Hunyuan OCR vs FireRed OCR - Which Handles Your Documents Better?
Side-by-side comparison of Hunyuan OCR and FireRed OCR on 50 scanned pages across 7 document types. Includes CER scores, latency, and routing recommendations from our first-party benchmark.
Best OCR for Scanned PDFs - 5 Models Tested on 50 Scanned Pages
We tested 5 OCR models on 50 scan-heavy pages across 7 document types. Here's which model to use for clean scans, degraded scans, tables, formulas, and mixed documents.
LLM vs OCR Is the Wrong Debate - Here's the Actual Taxonomy in 2026
Most "OCR models" in 2026 are multimodal LLMs. The real questions are specialist vs general-purpose, end-to-end vs hybrid pipeline, and accuracy vs hallucination risk. A four-tier taxonomy with a decision framework for production teams.
OmniDocBench Is Saturated - What Our 1,331-Page Benchmark Reveals About Real OCR Failures
OmniDocBench hits 94%+ accuracy across top models, but document parsing is far from solved. Our 1,331-page benchmark on scan-heavy PDFs exposes the failure modes that standard benchmarks miss: hallucinated text, broken table structures, spaced-letter artifacts, and blank-page blindness.