Models
29 models evaluated across 3 document AI benchmarks.
1
Nanonets OCR-3
Nanonets
Overall
85.9
2
GPT-5.4
OpenAI
Overall
83.5
3
Gemini-3-Pro
Overall
82.8
4
Gemini-3-Flash
Overall
82.0
5
Nanonets OCR2+
Nanonets
Overall
81.8
6
Gemini 3.1 Pro
Overall
81.6
7
GPT-5.2
OpenAI
Overall
81.5
8
Claude Sonnet 4.6
Anthropic
Overall
80.7
9
Claude Opus 4.6
Anthropic
Overall
80.4
10
Qwen3-VL-Plus
Alibaba
Overall
80.1
11
Qwen3-VL-235B
Alibaba
Overall
79.6
12
Qwen3.5-9B
Alibaba
Overall
76.7
13
GPT-5-Mini
OpenAI
Overall
75.2
14
Qwen3.5-4B
Alibaba
Overall
72.5
15
Mistral Small 4
Mistral AI
Overall
71.5
16
Claude Haiku 4.5
Anthropic
Overall
71.2
17
Ministral-8B
Mistral AI
Overall
69.5
18
GPT-4.1
OpenAI
Overall
68.0
19
GLM-OCR
Zhipu AI
Overall
64.2
20
Qwen3.5-2B
Alibaba
Overall
62.6
21
Qwen3.5-0.8B
Alibaba
Overall
57.8
22
GPT-5-Nano
OpenAI
Overall
54.8
23
Gemma-4-E4B-it
Overall
53.9
24
Llama-3.2-Vision-11B
Meta
Overall
50.8
25
Pixtral-12B
Mistral AI
Overall
46.5
26
Gemma-4-E2B-it
Overall
41.9
0
Gemma-3-12B-IT
Overall
0.0
0
Datalab Marker
Datalab
Overall
0.0
0
Qwen-VL-OCR
Alibaba
Overall
0.0