Skip to content

Benchmark

OlmOCR Bench

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

Models measured
33
Scores recorded
33
Sources
3
IQ weight
0.88

Score distribution

69.1%
Mean
75.3%
Median
15.3%
Std dev
47%
10th percentile
83.2%
90th percentile
20.6%
Min
20.6%86.3%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 0.88 for chat models, measured over 23 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
1LightOnOCR-3 4B· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claim86.3%
2Lightonocr 3 0.8b· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claim85.5%
3Jina OCR v1· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claim83.4%
4LightOnOCR-2-1B· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claim
Values from 2 sources
83.2%
5GPT 5.4· Nanonets IDP data81%
6PaddleOCR VL· OpenEvals Leaderboard data· Reported by the model's developer. The source relaying it did not verify it.80%
7Qianfan-OCR· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claim79.8%
8GPT 5.2· Nanonets IDP data79.1%
9Qwen3 VL Plus· Nanonets IDP data77.9%
10Gemini 3 Pro Preview· Nanonets IDP data77.7%
11Qwen3.5 9B· Nanonets IDP data77.2%
12Qwen3 VL 235B· Nanonets IDP data76.8%
13DeepSeek OCR 2· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claim
Values from 2 sources
76.3%
14DeepSeek-OCR· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claim
Values from 2 sources
75.7%
15Qwen3.5 4B· Nanonets IDP data75.4%
16Gemini 3 Flash· Nanonets IDP data75.3%
17GLM OCR· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claim
Values from 3 sources
75.2%
18Claude Opus 4.6· Nanonets IDP data74.1%
19Claude Sonnet 4.6· Nanonets IDP data73.9%
20Qwen3.5 2B· Nanonets IDP data71.9%
21GPT-5 mini· Nanonets IDP data69.9%
22Gemini 3.1 Pro Preview· Nanonets IDP data69.8%
23Mistral Small· Nanonets IDP data69.6%
24Qwen3.5 0.8B· Nanonets IDP data64.8%
25Claude Haiku 4.5· Nanonets IDP data61.2%
26Qwen VL OCR· Nanonets IDP data59%
27Ministral 8B· Nanonets IDP data58.7%
28GPT 4.1· Nanonets IDP data49.4%
29Gemma 4 E4B-it· Nanonets IDP data47%
30Pixtral 12B· Nanonets IDP data38.3%
31Gemma 4 E2B-it· Nanonets IDP data38.2%
32GPT-5 nano· Nanonets IDP data35.2%
33Gemma 3 12B IT· Nanonets IDP data20.6%

Data credits: Hugging Face official benchmark leaderboards; licence varies by benchmark · OpenEvals/leaderboard-data, MIT (licence)