Codesota · OCR · Benchmarks · OCRBench v2Home/OCR/Benchmarks/OCRBench v2
South China University of Technology

OCRBench v2.

Comprehensive benchmark evaluating 8 OCR capabilities across 23 tasks in 31 scenarios.

View on AlphaXiv ↗
§ 01 · Overall (Chinese)

Overall (Chinese).

Average score on Chinese private test set

Higher is better

#ModelScoreSource
★
TeleMM-2.0
OCRBench v2 ZH #1, official leaderboard 2026.03 (closed)
66.2OCRBench v2 official leaderboard
2
Qwen3.5-9B
OCRBench v2 ZH top open model, official leaderboard 2026.03
64.1OCRBench v2 official leaderboard
3
Gemini 3 Pro Preview
OCRBench v2 ZH, official leaderboard 2026.03
63.8OCRBench v2 official leaderboard
4
Qwen2.5-VL-72B
Fetched from CodeSOTA API on 2026-04-20
63.7codesota-api
5
gemini-25-pro
Fetched from CodeSOTA API on 2026-04-20
62.2codesota-api
6
MiniCPM-o-4.5
OCRBench v2 ZH, official leaderboard 2026.03
61.5OCRBench v2 official leaderboard
7
Qianfan-OCR
Fetched from CodeSOTA API on 2026-04-20
60.77codesota-api
8
intern-s1-pro
Mapped from PWC OCRBench v2 Chinese Score.; Reported in the Intern-S1-Pro paper and Hugging Face model card performance table as OCRBench V2 (ENG / CHN). OCRBench V2 is evaluated with the non-thinking configuration; scores are English 60.1 and Chinese 60.6.; PWC evaluation id 5083; paper: Intern-S1-Pro: Scientific Multimodal Foundation Model at Trillion Scale
60.6paperswithcode-public-api
9
minicpm-v-4.5-8b
Fetched from CodeSOTA API on 2026-04-20
58.8codesota-api
10
ovis2-5-9b
Mapped from PWC OCRBench v2 Chinese Score.; Table 6, OCR & chart; OCRBench v2 Chinese split. Source/provenance: Ovis2.5 Technical Report; source arXiv paper https://arxiv.org/abs/2508.11737; official HF model URL https://huggingface.co/AIDC-AI/Ovis2.5-9B.; PWC evaluation id 5587; paper: Ovis2.5 Technical Report
58paperswithcode-public-api
11
sail-vl2-8b
Fetched from CodeSOTA API on 2026-04-20
57.6codesota-api
12
claude-3.5-sonnet
Fetched from CodeSOTA API on 2026-04-20
48.4codesota-api
13
InternVL2.5-78B
Fetched from CodeSOTA API on 2026-04-20
46.2codesota-api
14
Qwen2-VL-72B
Fetched from CodeSOTA API on 2026-04-20
46.1codesota-api
15
gpt-4o-2024
Fetched from CodeSOTA API on 2026-04-20
45.7codesota-api
§ 02 · Overall (English)

Overall (English).

Average score on English private test set

Higher is better

#ModelScoreSource
★
KDL Frontier
OCRBench v2 EN #1, official leaderboard 2026.03 (closed)
68.1OCRBench v2 official leaderboard
2
Nemotron-3-Nano-Omni-30B
OCRBench v2 EN top open model, official leaderboard 2026.03
65.8OCRBench v2 official leaderboard
3
ovis2-5-9b
Mapped from PWC OCRBench v2 English Score.; Table 6, OCR & chart; OCRBench v2 English split. Source/provenance: Ovis2.5 Technical Report; source arXiv paper https://arxiv.org/abs/2508.11737; official HF model URL https://huggingface.co/AIDC-AI/Ovis2.5-9B.; PWC evaluation id 5586; paper: Ovis2.5 Technical Report
63.4paperswithcode-public-api
4
Gemini 3 Pro Preview
OCRBench v2 EN, official leaderboard 2026.03
63.4OCRBench v2 official leaderboard
5
seed-1.6-vision
Fetched from CodeSOTA API on 2026-04-20
62.2codesota-api
6
TeleMM-2.0
OCRBench v2 EN, official leaderboard 2026.03 (closed)
61.8OCRBench v2 official leaderboard
7
Qwen2.5-VL-72B
Fetched from CodeSOTA API on 2026-04-20
61.5codesota-api
8
qwen3-omni-30b
Fetched from CodeSOTA API on 2026-04-20
61.3codesota-api
9
nemotron-nano-v2-vl
Fetched from CodeSOTA API on 2026-04-20
61.2codesota-api
10
intern-s1-pro
Mapped from PWC OCRBench v2 English Score.; Reported in the Intern-S1-Pro paper and Hugging Face model card performance table as OCRBench V2 (ENG / CHN). OCRBench V2 is evaluated with the non-thinking configuration; scores are English 60.1 and Chinese 60.6.; PWC evaluation id 5083; paper: Intern-S1-Pro: Scientific Multimodal Foundation Model at Trillion Scale
60.1paperswithcode-public-api
11
gemini-25-pro
Fetched from CodeSOTA API on 2026-04-20
59.3codesota-api
12
llama-3.1-nemotron-nano-vl-8b
Fetched from CodeSOTA API on 2026-04-20
56.4codesota-api
13
Qianfan-OCR
Fetched from CodeSOTA API on 2026-04-20
56codesota-api
14
gpt-4o
Fetched from CodeSOTA API on 2026-04-20
55.5codesota-api
15
ovis2.5-8b
Fetched from CodeSOTA API on 2026-04-20
54.1codesota-api
16
gemini-1.5-pro
Fetched from CodeSOTA API on 2026-04-20
51.6codesota-api
17
sail-vl2-8b
Fetched from CodeSOTA API on 2026-04-20
49.3codesota-api
18
minicpm-v-4.5-8b
Fetched from CodeSOTA API on 2026-04-20
48.4codesota-api
19
Qwen2-VL-72B
Fetched from CodeSOTA API on 2026-04-20
47.8codesota-api
20
gpt-4o-2024
Fetched from CodeSOTA API on 2026-04-20
47.6codesota-api
21
claude-3.5-sonnet
Fetched from CodeSOTA API on 2026-04-20
47.5codesota-api
22
internvl3.5-14b
Fetched from CodeSOTA API on 2026-04-20
47.1codesota-api
23
step-1v
Fetched from CodeSOTA API on 2026-04-20
46.8codesota-api
24
InternVL2.5-78B
Fetched from CodeSOTA API on 2026-04-20
45codesota-api
25
grok4
Fetched from CodeSOTA API on 2026-04-20
45codesota-api
26
gpt-4o-mini
Fetched from CodeSOTA API on 2026-04-20
44.1codesota-api
27
claude-sonnet-4
Fetched from CodeSOTA API on 2026-04-20
42.4codesota-api
28
qwen2.5-vl-7b
Fetched from CodeSOTA API on 2026-04-20
41.8codesota-api
29
deepseek-vl2-small
Fetched from CodeSOTA API on 2026-04-20
41codesota-api
30
pixtral-12b
Fetched from CodeSOTA API on 2026-04-20
38.4codesota-api
31
phi-4-multimodal
Fetched from CodeSOTA API on 2026-04-20
38.1codesota-api
32
glm-4v-9b
Fetched from CodeSOTA API on 2026-04-20
37.1codesota-api
33
molmo-7b
Fetched from CodeSOTA API on 2026-04-20
33.9codesota-api
34
llava-ov-7b
Fetched from CodeSOTA API on 2026-04-20
33.7codesota-api
35
idefics3-8b
Fetched from CodeSOTA API on 2026-04-20
26codesota-api
36
mistral-ocr-2512
Fetched from CodeSOTA API on 2026-04-20
25.2codesota-api
37
docowl2
Fetched from CodeSOTA API on 2026-04-20
23.4codesota-api
§ 03 · overall-zh-public

overall-zh-public.

Higher is better

#ModelScoreSource
★
InternVL3-14B
Fetched from CodeSOTA API on 2026-04-20
55.7codesota-api
2
Qwen2.5-VL-7B
Fetched from CodeSOTA API on 2026-04-20
55.6codesota-api
3
Ovis2-8B
Fetched from CodeSOTA API on 2026-04-20
49.2codesota-api
4
Gemini 1.5 Pro
Fetched from CodeSOTA API on 2026-04-20
43.1codesota-api
5
DeepSeek-VL2-Small
Fetched from CodeSOTA API on 2026-04-20
42.7codesota-api
6
Step-1V
Fetched from CodeSOTA API on 2026-04-20
42.6codesota-api
7
MiniCPM-o-2.6
Fetched from CodeSOTA API on 2026-04-20
41.1codesota-api
8
Claude 3.5 Sonnet
Fetched from CodeSOTA API on 2026-04-20
39.6codesota-api
9
GLM-4V-9B
Fetched from CodeSOTA API on 2026-04-20
36.6codesota-api
10
GPT-4o
Fetched from CodeSOTA API on 2026-04-20
32.2codesota-api
11
LLaVA-OneVision-7B
Fetched from CodeSOTA API on 2026-04-20
17.8codesota-api
12
TextMonkey
Fetched from CodeSOTA API on 2026-04-20
15.8codesota-api
13
Pixtral-12B
Fetched from CodeSOTA API on 2026-04-20
14.6codesota-api
14
Monkey
Fetched from CodeSOTA API on 2026-04-20
13.1codesota-api
15
Molmo-7B
Fetched from CodeSOTA API on 2026-04-20
12.8codesota-api
16
Cambrian-1-8B
Fetched from CodeSOTA API on 2026-04-20
9.90codesota-api
17
LLaVA-NeXT-8B
Fetched from CodeSOTA API on 2026-04-20
9.10codesota-api
§ 04 · overall-en-public

overall-en-public.

Higher is better

#ModelScoreSource
★
InternVL3-14B
Fetched from CodeSOTA API on 2026-04-20
52.6codesota-api
2
Gemini 1.5 Pro
Fetched from CodeSOTA API on 2026-04-20
51.9codesota-api
3
Ovis2-8B
Fetched from CodeSOTA API on 2026-04-20
47.7codesota-api
4
Qwen2.5-VL-7B
Fetched from CodeSOTA API on 2026-04-20
46.7codesota-api
5
Step-1V
Fetched from CodeSOTA API on 2026-04-20
46.7codesota-api
6
GPT-4o
Fetched from CodeSOTA API on 2026-04-20
46.5codesota-api
7
Claude 3.5 Sonnet
Fetched from CodeSOTA API on 2026-04-20
45.2codesota-api
8
MiniCPM-o-2.6
Fetched from CodeSOTA API on 2026-04-20
45.1codesota-api
9
DeepSeek-VL2-Small
Fetched from CodeSOTA API on 2026-04-20
43.3codesota-api
10
GLM-4V-9B
Fetched from CodeSOTA API on 2026-04-20
42.6codesota-api
11
Pixtral-12B
Fetched from CodeSOTA API on 2026-04-20
40.3codesota-api
12
LLaVA-OneVision-7B
Fetched from CodeSOTA API on 2026-04-20
36.4codesota-api
13
Cambrian-1-8B
Fetched from CodeSOTA API on 2026-04-20
34.7codesota-api
14
Molmo-7B
Fetched from CodeSOTA API on 2026-04-20
34.5codesota-api
15
LLaVA-NeXT-8B
Fetched from CodeSOTA API on 2026-04-20
31.5codesota-api
16
TextMonkey
Fetched from CodeSOTA API on 2026-04-20
23.9codesota-api
17
Monkey
Fetched from CodeSOTA API on 2026-04-20
23.1codesota-api
§ Related · Explore

More OCR content.

Verified Model Reviews
Comparisons & Guides
View all OCR benchmarks →← Back to All Benchmarks