Historical registry · Presentation reviewed 7 October 2026
OCR evidence by benchmark
The former “power ranking” averaged percentiles across different tasks and evaluation populations. It did not establish which model is best for a document workflow, so we removed the overall score and winner framing.
Tables below separate benchmark and metric. Rows are alphabetical because matching dataset names alone do not establish matched versions, prompts or harnesses. Imported registry flags do not mean CodeSOTA reproduced the run. Missing evidence metadata remains visible.
OmniDocBench composite
omnidocbench / composite · 40 reported observations · inspect source definition before comparing
| Model | Reported score | Result date | Import date | Version / protocol | Source |
|---|---|---|---|---|---|
| clearOCR | 31.70 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| DeepSeek-OCR | 86.46 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| DeepSeek-OCR-2 | 91.09 | 2026-01-28 | 2026-05-15 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| Dolphin | 74.67 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| Dolphin-1.5 | 85.06 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| Dolphin-v2 | 89.78 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| dots.ocr 3B | 88.41 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| Falcon-OCR | 88.64 | 2026-03-28 | 2026-05-15 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| FireRed-OCR-2B | 92.94 | 2026-03-02 | 2026-05-15 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| Gemini 2.5 Pro | 88.03 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| Gemini 3 Pro | 90.33 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| GLM-OCR | 94.62 | 2026-03-11 | 2026-05-15 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| GPT-4o | 75.02 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| InternVL3-76B | 80.33 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| InternVL3.5-241B | 82.67 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| Kimi K2.5 | 88.80 | 2026-02-02 | 2026-05-15 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| Marker 1.8.2 | 71.30 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| MinerU 2.5 | 90.67 | 2025-09-26 | 2026-05-15 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| MinerU2-pipeline | 75.51 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| MinerU2-VLM | 85.56 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| Mistral OCR 3 | 79.75 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| mistral-ocr-2512 | 79.75 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| MonkeyOCR-3B | 87.13 | 2025-06-05 | 2026-05-15 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| MonkeyOCR-pro-1.2B | 86.96 | 2025-06-05 | 2026-05-15 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| MonkeyOCR-pro-3B | 88.85 | 2025-06-05 | 2026-05-15 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| Nanonets-OCR-s | 85.59 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| OCRFlux-3B | 74.82 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| OCRVerse 4B | 88.56 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| olmOCR | 81.79 | 2025-02-25 | 2026-05-15 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| olmOCR-7B | 81.79 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| PaddleOCR-VL | 92.86 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| PaddleOCR-VL | 92.56 | 2025-10-16 | 2026-05-15 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| PaddleOCR-VL 0.9B | 92.56 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| PaddleOCR-VL-1.5 | 94.50 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| POINTS-Reader | 80.98 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| PP-StructureV3 | 86.73 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| Qianfan-OCR | 93.12 | 2026-03-11 | 2026-05-15 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| Qwen2.5-VL | 87.02 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| Qwen3-VL-235B | 89.15 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| Qwen3.5-397B-A17B | 90.80 | 2026-02-16 | 2026-05-18 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
OCRBench score
ocrbench / score · 30 reported observations · inspect source definition before comparing
| Model | Reported score | Result date | Import date | Version / protocol | Source |
|---|---|---|---|---|---|
| dots.mocr | 860 | 2026-03-13 | 2026-05-18 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| HunyuanOCR (1B) | 860 | 2025-11-24 | 2026-05-18 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| Infinity-Parser2-Pro | 862 | 2026-05-11 | 2026-05-18 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| InternVL3-78B | 906 | 2025-04-14 | 2026-05-14 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| Kimi K2.5 | 923 | 2026-02-02 | 2026-05-13 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| Kimi-VL-A3B-Instruct | 867 | 2025-04-10 | 2026-05-15 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| Kimi-VL-A3B-Thinking-2506 | 869 | 2025-04-10 | 2026-05-15 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| MiniCPM-Llama3-V 2.5 | 725 | 2024-08-03 | 2026-05-18 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| MiniCPM-o 4.5-Instruct | 876 | 2026-04-30 | 2026-05-15 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| MiniCPM-V 4.6-Thinking (16x) | 831 | 2026-05-13 | 2026-05-18 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| MiniMax-VL-01 | 865 | 2025-01-14 | 2026-05-15 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| Ovis2.5-9B | 879 | 2025-08-15 | 2026-05-18 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| Qianfan-OCR | 880 | 2026-03-11 | 2026-05-15 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| Qwen2-VL-2B | 809 | 2024-09-18 | 2026-05-18 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| Qwen2-VL-72B | 877 | 2024-09-18 | 2026-05-15 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| Qwen2-VL-7B | 866 | 2024-09-18 | 2026-05-15 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| Qwen2.5-VL-3B | 797 | 2025-02-19 | 2026-05-18 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| Qwen2.5-VL-72B | 885 | 2025-02-19 | 2026-05-15 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| Qwen2.5-VL-7B | 864 | 2025-02-19 | 2026-05-18 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| Qwen3-VL-235B-A22B-Instruct | 920 | 2025-11-26 | 2026-05-14 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| Qwen3-VL-235B-A22B-Thinking | 875 | 2025-11-26 | 2026-05-15 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| Qwen3-VL-8B-Instruct | 896 | 2025-11-26 | 2026-05-14 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| Qwen3.5-397B-A17B | 931 | 2026-02-16 | 2026-05-18 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| Qwen3.5-Omni-Plus | 913 | 2026-04-17 | 2026-05-18 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| Qwen3.6-27B | 894 | 2026-04-21 | 2026-05-18 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| Qwen3.6-35B-A3B | 900 | 2026-04-21 | 2026-05-18 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| SenseNova-U1-A3B-MoT | 919 | 2026-05-12 | 2026-05-18 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| VideoLLaMA3 2B | 779 | 2025-01-22 | 2026-05-18 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| VideoLLaMA3 7B | 828 | 2025-01-22 | 2026-05-18 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| ZAYA1-VL-8B | 798 | 2026-05-08 | 2026-05-18 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
OCRBench v2 English
ocrbench-v2 / overall-en-private · 33 reported observations · inspect source definition before comparing
| Model | Reported score | Result date | Import date | Version / protocol | Source |
|---|---|---|---|---|---|
| Claude Sonnet 4 | 42.40 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| claude-3.5-sonnet | 47.50 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| deepseek-vl2-small | 41.00 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| docowl2 | 23.40 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| Gemini 2.5 Pro | 59.30 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| gemini-1.5-pro | 51.60 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| glm-4v-9b | 37.10 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| GPT-4o | 55.50 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| GPT-4o Mini | 44.10 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| gpt-4o-2024 | 47.60 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| grok4 | 45.00 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| idefics3-8b | 26.00 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| Intern-S1-Pro | 60.10 | 2026-03-26 | 2026-05-18 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| InternVL2.5-78B | 45.00 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| internvl3.5-14b | 47.10 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| llama-3.1-nemotron-nano-vl-8b | 56.40 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| llava-ov-7b | 33.70 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| minicpm-v-4.5-8b | 48.40 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| mistral-ocr-2512 | 25.20 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| molmo-7b | 33.90 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| Nemotron Nano V2 VL | 61.20 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| ovis2.5-8b | 54.10 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| Ovis2.5-9B | 63.40 | 2025-08-15 | 2026-05-18 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| phi-4-multimodal | 38.10 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| pixtral-12b | 38.40 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| Qianfan-OCR | 56.00 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| Qwen2-VL-72B | 47.80 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| Qwen2.5-VL-72B | 61.50 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| qwen2.5-vl-7b | 41.80 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| Qwen3-Omni-30B | 61.30 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| sail-vl2-8b | 49.30 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| Seed1.6-vision | 62.20 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| step-1v | 46.80 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
OCRBench v2 Chinese
ocrbench-v2 / overall-zh-private · 11 reported observations · inspect source definition before comparing
| Model | Reported score | Result date | Import date | Version / protocol | Source |
|---|---|---|---|---|---|
| claude-3.5-sonnet | 48.40 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| Gemini 2.5 Pro | 62.20 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| gpt-4o-2024 | 45.70 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| Intern-S1-Pro | 60.60 | 2026-03-26 | 2026-05-18 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| InternVL2.5-78B | 46.20 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| minicpm-v-4.5-8b | 58.80 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| Ovis2.5-9B | 58.00 | 2025-08-15 | 2026-05-18 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| Qianfan-OCR | 60.77 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| Qwen2-VL-72B | 46.10 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| Qwen2.5-VL-72B | 63.70 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| sail-vl2-8b | 57.60 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
olmOCR pass rate
olmocr-bench / pass-rate · 33 reported observations · inspect source definition before comparing
| Model | Reported score | Result date | Import date | Version / protocol | Source |
|---|---|---|---|---|---|
| Chandra | 83.10 | 2025-10-21 | 2026-05-15 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| Chandra 2 | 85.90 | 2026-03-16 | 2026-05-15 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| Chandra v0.1.0 | 83.10 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| DeepSeek OCR | 75.70 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| DeepSeek-OCR | 75.70 | 2025-10-21 | 2026-05-15 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| DeepSeek-OCR-2 | 76.30 | 2026-01-28 | 2026-05-15 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| dots.mocr | 83.90 | 2026-03-13 | 2026-05-15 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| dots.ocr | 79.10 | 2025-12-02 | 2026-05-15 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| dots.ocr 3B | 79.10 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| Falcon-OCR | 80.30 | 2026-03-28 | 2026-05-15 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| FireRed-OCR | 70.20 | 2026-03-02 | 2026-05-15 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| Gemini Flash 2 | 63.80 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| GLM-OCR | 75.20 | 2026-03-11 | 2026-05-15 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| GPT-4o (Anchored) | 69.90 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| Infinity-Parser 7B | 82.50 | 2025-06-01 | 2026-05-15 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| Infinity-Parser2-Pro | 87.60 | 2026-05-11 | 2026-05-15 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| LightOnOCR-1B-1025 | 76.10 | 2026-01-20 | 2026-05-15 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| LightOnOCR-2-1B | 83.20 | 2026-01-20 | 2026-05-15 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| Marker 1.10.0 | 76.50 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| Marker 1.10.1 | 76.10 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| MinerU 2.5 | 75.20 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| MinerU 2.5 | 77.50 | 2025-09-26 | 2026-05-15 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| Mistral OCR 2 | 72.00 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| Mistral OCR 3 | 78.00 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| MonkeyOCR-pro-3B | 75.80 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| Nanonets OCR2 3B | 69.50 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| olmOCR | 75.50 | 2025-02-25 | 2026-05-15 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| olmOCR v0.4.0 | 82.40 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| olmOCR-2-7B-1025 (7B) | 82.40 | 2025-10-22 | 2026-05-15 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| PaddleOCR-VL | 80.00 | 2025-10-16 | 2026-05-15 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| PaddleOCR-VL-1.5 | 79.10 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| Qianfan-OCR | 79.80 | 2026-03-11 | 2026-05-15 | not recorded / not recorded | paperswithcode-public-api Inspect report ↗ |
| Qwen3-VL-4B | 79.20 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
CC-OCR multi-scene F1
cc-ocr / multi-scene-f1 · 5 reported observations · inspect source definition before comparing
| Model | Reported score | Result date | Import date | Version / protocol | Source |
|---|---|---|---|---|---|
| Claude 3.5 Sonnet | 72.87 | not recorded | 2026-04-20 | not recorded / not recorded | src Source URL not recorded |
| Gemini 1.5 Pro | 83.25 | not recorded | 2026-04-20 | not recorded / not recorded | src Source URL not recorded |
| GPT-4o | 76.40 | not recorded | 2026-04-20 | not recorded / not recorded | src Source URL not recorded |
| InternVL2-76B | 76.92 | not recorded | 2026-04-20 | not recorded / not recorded | src Source URL not recorded |
| Qwen2-VL 72B | 77.95 | not recorded | 2026-04-20 | not recorded / not recorded | src Source URL not recorded |
MME-VideoOCR total accuracy
mme-videoocr / total-accuracy · 6 reported observations · inspect source definition before comparing
| Model | Reported score | Result date | Import date | Version / protocol | Source |
|---|---|---|---|---|---|
| Gemini 1.5 Pro | 64.90 | not recorded | 2026-04-20 | not recorded / not recorded | src Source URL not recorded |
| Gemini 2.5 Pro | 73.70 | not recorded | 2026-04-20 | not recorded / not recorded | src Source URL not recorded |
| GPT-4o | 66.40 | not recorded | 2026-04-20 | not recorded / not recorded | src Source URL not recorded |
| InternVL3-78B | 67.20 | not recorded | 2026-04-20 | not recorded / not recorded | src Source URL not recorded |
| Qwen2.5-VL 32B | 61.00 | not recorded | 2026-04-20 | not recorded / not recorded | src Source URL not recorded |
| Qwen2.5-VL 72B | 69.00 | not recorded | 2026-04-20 | not recorded / not recorded | src Source URL not recorded |
KITAB character error rate
kitab-bench / cer · 8 reported observations · lower score is better within a matched protocol
| Model | Reported score | Result date | Import date | Version / protocol | Source |
|---|---|---|---|---|---|
| AIN 7B | 0.20 | not recorded | 2026-04-20 | not recorded / not recorded | src Source URL not recorded |
| Azure OCR | 0.52 | not recorded | 2026-04-20 | not recorded / not recorded | src Source URL not recorded |
| EasyOCR | 0.58 | not recorded | 2026-04-20 | not recorded / not recorded | src Source URL not recorded |
| Gemini 2.0 Flash | 0.13 | not recorded | 2026-04-20 | not recorded / not recorded | src Source URL not recorded |
| GPT-4o | 0.31 | not recorded | 2026-04-20 | not recorded / not recorded | src Source URL not recorded |
| GPT-4o Mini | 0.43 | not recorded | 2026-04-20 | not recorded / not recorded | src Source URL not recorded |
| PaddleOCR | 0.79 | not recorded | 2026-04-20 | not recorded / not recorded | src Source URL not recorded |
| Tesseract | 0.54 | not recorded | 2026-04-20 | not recorded / not recorded | src Source URL not recorded |
Thai OCR TED score
thaiocrbench / ted-score · 5 reported observations · inspect source definition before comparing
| Model | Reported score | Result date | Import date | Version / protocol | Source |
|---|---|---|---|---|---|
| Claude Sonnet 4 | 0.84 | not recorded | 2026-04-20 | not recorded / not recorded | src Source URL not recorded |
| Gemini 2.5 Pro | 0.77 | not recorded | 2026-04-20 | not recorded / not recorded | src Source URL not recorded |
| InternVL3 14B | 0.76 | not recorded | 2026-04-20 | not recorded / not recorded | src Source URL not recorded |
| Qwen2.5-VL 32B | 0.77 | not recorded | 2026-04-20 | not recorded / not recorded | src Source URL not recorded |
| Qwen2.5-VL 72B | 0.72 | not recorded | 2026-04-20 | not recorded / not recorded | src Source URL not recorded |
ParseBench accuracy
parsebench / accuracy · 14 reported observations · inspect source definition before comparing
| Model | Reported score | Result date | Import date | Version / protocol | Source |
|---|---|---|---|---|---|
| Anthropic Haiku 4.5 | 45.20 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| AWS Textract | 47.90 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| Azure Document Intelligence | 59.60 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| Docling | 50.60 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| Dots OCR 1.5 | 55.80 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| Extend | 55.80 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| Google Cloud Document AI | 50.40 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| Google Gemini 3 Flash | 71.00 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| LandingAI | 45.20 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| LlamaParse Agentic | 84.90 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| LlamaParse Cost Effective | 71.90 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| OpenAI GPT-5 Mini | 46.80 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| Qwen 3 VL | 62.00 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
| Reducto | 67.80 | not recorded | 2026-04-20 | not recorded / not recorded | codesota-api Inspect report ↗ |
Use the evidence for a shortlist
Match output, language, deployment and budget first. Then test candidate checkpoints on the same held-out documents. Publish benchmark version, split, prompts, normalization, outputs and run artifacts before claiming a reproducible ranking.