Skip to content

Benchmark

OmniDocBench

Vision benchmarks measure how well models understand images. Scores are percentages.

Models measured
34
Scores recorded
34
Sources
3
IQ weight
0.65

Score distribution

78.1%
Mean
85.3%
Median
17%
Std dev
47.3%
10th percentile
92.9%
90th percentile
34.1%
Min
34.1%96.3%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 0.65 for chat models, measured over 27 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 data96.3%
2GLM OCR· OmniDocBench data
Values from 2 sources
95.2%
3PaddleOCR VL· OmniDocBench data94.2%
4Qianfan-OCR· OmniDocBench data93.9%
5Gemini 3 Pro Preview· OmniDocBench data
Values from 2 sources
92.9%
6Inkling 2 Mini· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim92.8%
7Gemini 3 Flash· OmniDocBench data
Values from 2 sources
92.6%
8Kimi K3· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim91.1%
9DeepSeek OCR 2· OmniDocBench data90.3%
10HunyuanOCR· OmniDocBench data90%
11Qwen3 VL 235B· OmniDocBench data
Values from 2 sources
89.8%
12GPT 5.2· Nanonets IDP data
Values from 2 sources
88%
13MiMo V2.5· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim87.2%
14Claude Sonnet 4.6· Nanonets IDP data86.9%
15Claude Opus 4.6· Nanonets IDP data85.9%
16Gemini 3.1 Pro Preview· Nanonets IDP data85.3%
17GPT 5.4· Nanonets IDP data85.3%
18Kimi K2.5· OmniDocBench data84.5%
19Qwen3 VL Plus· Nanonets IDP data82.5%
20GPT-5 mini· Nanonets IDP data82.5%
21GPT 4.1· Nanonets IDP data79.9%
22Claude Haiku 4.5· Nanonets IDP data79.6%
23Ministral 8B· Nanonets IDP data78.3%
24Qwen3.5 9B· Nanonets IDP data76.7%
25Mistral Small· Nanonets IDP data76.4%
26Qwen3.5 4B· Nanonets IDP data67.6%
27GPT-5 nano· Nanonets IDP data63.4%
28Gemma 4 E4B-it· Nanonets IDP data59.7%
29Qwen3.5 2B· Nanonets IDP data48.7%
30Qwen3.5 0.8B· Nanonets IDP data47.3%
31Gemma 3 12B IT· Nanonets IDP data44.6%
32Gemma 4 E2B-it· Nanonets IDP data43.3%
33Pixtral 12B· Nanonets IDP data42.3%
34Qwen VL OCR· Nanonets IDP data34.1%

Data credits: OmniDocBench (OpenDataLab), Apache-2.0 (licence) · Data by LLM Stats