Vision & documents

OCRBench V2 leaderboard

Name: benchgap: LLM benchmark scores, measured and estimated
Creator: benchgap
License: https://creativecommons.org/licenses/by-nc/4.0/

As of 2026-10-29, the highest measured score on OCRBench V2 is 74.2% by Qwen3.8 Max. 53 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	Qwen3.8-Flash-Next	74.4%	estimated ± 3.6 pp, low confidence
2	GPT-6.1 Terra	74.4%	estimated ± 4.4 pp, low confidence
3	Qwen3.8 Max	74.2%	measured
4	Kimi K3.1 Pro	74.1%	estimated ± 4.4 pp, low confidence
5	Gemini 3.1 Pro	74.0%	estimated ± 4.4 pp, low confidence
6	Gemini 3.5 Flash	73.9%	estimated ± 4.4 pp, low confidence
7	GPT-5.6 Sol	73.8%	estimated ± 4.4 pp, low confidence
8	Kimi K3	73.4%	estimated ± 4.4 pp, medium confidence
9	Seed 2.1 Pro	73.4%	estimated ± 4.4 pp, medium confidence
10	GPT-5.4	73.3%	estimated ± 4.4 pp, medium confidence
11	GPT-5.5	73.3%	estimated ± 4.4 pp, medium confidence
12	GPT-5.6 Terra	73.2%	estimated ± 4.4 pp, medium confidence
13	Muse Spark	73.1%	estimated ± 4.4 pp, medium confidence
14	Qwen3.8-Omni-Flash	73.1%	estimated ± 3.6 pp, medium confidence
15	Seed 2.1 Turbo	73.1%	estimated ± 4.4 pp, medium confidence
16	Kimi K2.6	72.9%	estimated ± 4.4 pp, medium confidence
17	dots3-note Preview	72.8%	estimated ± 4.4 pp, medium confidence
18	Qwen3.6 Plus	72.7%	estimated ± 4.4 pp, medium confidence
19	Kimi K2.5 (Reasoning)	72.6%	estimated ± 4.4 pp, medium confidence
20	GPT-5.6 Luna	72.6%	estimated ± 4.4 pp, medium confidence
21	Grok 4.3	72.5%	estimated ± 4.4 pp, medium confidence
22	MiniMax M3	72.5%	estimated ± 4.4 pp, medium confidence
23	Pareto 26.9	72.5%	estimated ± 4.4 pp, medium confidence
24	MiMo-V2.5	72.5%	estimated ± 4.4 pp, medium confidence
25	Claude Opus 4.6	72.3%	estimated ± 4.4 pp, medium confidence
26	Gemma 4 31B	72.2%	estimated ± 4.4 pp, medium confidence
27	Qwen3.5-122B-A10B	72.2%	estimated ± 4.4 pp, medium confidence
28	GPT-5.4 mini	72.1%	estimated ± 4.4 pp, medium confidence
29	Step 5 Preview	71.9%	estimated ± 4.4 pp, medium confidence
30	Grok 4.20	71.7%	estimated ± 4.4 pp, medium confidence
31	Qwen3.5-35B-A3B	71.7%	estimated ± 4.4 pp, medium confidence
32	Qwen3.5-27B	71.7%	estimated ± 4.4 pp, medium confidence
33	Inkling-Small	71.4%	estimated ± 4.4 pp, medium confidence
34	Muse Glimmer 30B	71.4%	estimated ± 4.4 pp, medium confidence
35	Gemma 4 26B A4B	71.3%	estimated ± 4.4 pp, medium confidence
36	Inkling	71.2%	estimated ± 4.4 pp, medium confidence
37	Interfaze Beta	70.7%	measured
38	Qwen3.7 Plus	70.7%	measured
39	Qwen3.8-27B	70.0%	estimated ± 3.6 pp, medium confidence
40	Gemma 4 12B	69.9%	estimated ± 4.4 pp, medium confidence
41	Qwen3.6-35B-A3B	69.0%	estimated ± 3.6 pp, medium confidence
42	GPT-5.4 nano	68.9%	estimated ± 4.4 pp, medium confidence
43	Command A+	67.8%	estimated ± 4.4 pp, medium confidence
44	Qwen3.6-27B	67.0%	estimated ± 3.6 pp, medium confidence
45	Qwen3.5 397B	66.6%	estimated ± 3.6 pp, medium confidence
46	Gemini 3 Pro	65.6%	estimated ± 3.6 pp, medium confidence
47	GPT-5.2	65.6%	estimated ± 3.6 pp, medium confidence
48	Gemma 4 E4B	63.5%	estimated ± 4.4 pp, medium confidence
49	LFM3-VL-2B	63.5%	measured
50	Qwen2.5-VL-32B	62.0%	estimated ± 4.4 pp, medium confidence
51	Mistral Small 3.1	61.9%	estimated ± 4.4 pp, medium confidence
52	Kimi K2.5	61.7%	estimated ± 3.6 pp, medium confidence
53	Gemma 4 E2B	59.2%	estimated ± 4.4 pp, medium confidence
54	Ternary Bonsai 2 27B	56.9%	measured
55	Claude Opus 4.5	55.0%	estimated ± 3.6 pp, medium confidence
56	LFM2.5-VL-3B	47.5%	measured
57	ZAYA1-VL-8B	34.6%	estimated ± 3.6 pp, low confidence
58	North Micro Vision Instruct	29.9%	estimated ± 3.6 pp, low confidence
59	LFM2.5-VL-450M	23.5%	estimated ± 3.6 pp, low confidence

OCRBench V2 leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General