Skip to content

Benchmark

OmniDocBench - Text Edit

OCR benchmarks measure how well models read text in images and documents. Scores are percentages. Lower is better.

Models measured
31
Scores recorded
31
Sources
2
IQ weight
1.76

Score distribution

21.6%
Mean
15.1%
Median
20%
Std dev
5%
10th percentile
58.3%
90th percentile
3.3%
Min
3.3%82.3%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.76 for chat models, measured over 24 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
1PaddleOCR-VL-1.6· OmniDocBench data3.3%
2PaddleOCR VL· OmniDocBench data4%
3Qianfan-OCR· OmniDocBench data4%
4GLM OCR· OmniDocBench data
Values from 2 sources
4.4%
5DeepSeek OCR 2· OmniDocBench data5%
6Qwen3 VL 235B· OmniDocBench data
Values from 2 sources
6.3%
7Gemini 3 Pro Preview· OmniDocBench data
Values from 2 sources
6.4%
8Gemini 3 Flash· OmniDocBench data
Values from 2 sources
6.6%
9Gemini 3.1 Pro Preview· Nanonets IDP data8.2%
10HunyuanOCR· OmniDocBench data8.8%
11GPT 5.4· Nanonets IDP data8.9%
12Kimi K2.5· OmniDocBench data10.7%
13GPT 5.2· Nanonets IDP data
Values from 2 sources
11.1%
14GPT-5 mini· Nanonets IDP data13.8%
15Claude Opus 4.6· Nanonets IDP data15.1%
16Ministral 8B· Nanonets IDP data15.7%
17Qwen3 VL Plus· Nanonets IDP data15.7%
18Claude Sonnet 4.6· Nanonets IDP data16.5%
19GPT 4.1· Nanonets IDP data16.7%
20Claude Haiku 4.5· Nanonets IDP data22.4%
21Mistral Small· Nanonets IDP data24.2%
22Qwen3.5 9B· Nanonets IDP data25.3%
23Qwen3.5 4B· Nanonets IDP data29.2%
24GPT-5 nano· Nanonets IDP data31.9%
25Gemma 4 E4B-it· Nanonets IDP data36.7%
26Gemma 3 12B IT· Nanonets IDP data47.6%
27Gemma 4 E2B-it· Nanonets IDP data55.3%
28Qwen3.5 0.8B· Nanonets IDP data58.3%
29Qwen3.5 2B· Nanonets IDP data62.1%
30Pixtral 12B· Nanonets IDP data64.1%
31Qwen VL OCR· Nanonets IDP data82.3%

Data credits: OmniDocBench (OpenDataLab), Apache-2.0 (licence)