OCRBench

Name: OCRBench Leaderboard — AI Model Scores
Creator: LLM Stats
License: https://llm-stats.com/legal/terms-of-service

Paper

Progress Over Time

Interactive timeline showing model performance evolution on OCRBench

State-of-the-art frontier

Open

Proprietary

OCRBench Leaderboard

24 models

			Context	Cost
1	Kimi K2.5 Moonshot AI	1.0T	—	—
2	Qwen3.5-122B-A10B Alibaba Cloud / Qwen Team	122B	—	—
3	Qwen3 VL 235B A22B Instruct Alibaba Cloud / Qwen Team	236B	—	—
4	Qwen3.5-35B-A3B Alibaba Cloud / Qwen Team	35B	—	—
5	Qwen3 VL 30B A3B Instruct Alibaba Cloud / Qwen Team	31B	—	—
6	Qwen3 VL 8B Instruct Alibaba Cloud / Qwen Team	9B	—	—
7	Qwen3 VL 32B Instruct Alibaba Cloud / Qwen Team	33B	—	—
8	Qwen3.5-27B Alibaba Cloud / Qwen Team	27B	262K	$0.30 / $2.40
8	Qwen3.6-27B Alibaba Cloud / Qwen Team	28B	262K	$0.60 / $3.60
10	Qwen2.5 VL 72B Instruct Alibaba Cloud / Qwen Team	72B	—	—
11	Qwen3 VL 4B Instruct Alibaba Cloud / Qwen Team	4B	262K	$0.10 / $0.60
12	Qwen2-VL-72B-Instruct Alibaba Cloud / Qwen Team	73B	—	—
13	Qwen3 VL 235B A22B Thinking Alibaba Cloud / Qwen Team	236B	—	—
14	Qwen2.5 VL 7B Instruct Alibaba Cloud / Qwen Team	8B	—	—
15	Qwen3 VL 32B Thinking Alibaba Cloud / Qwen Team	33B	—	—
16	Phi-4-multimodal-instruct Microsoft	6B	—	—
17	LFM2.5-VL-3B Liquid AI	3B	—	—
18	Qwen3 VL 30B A3B Thinking Alibaba Cloud / Qwen Team	31B	—	—
19	DeepSeek VL2 Small DeepSeek	16B	—	—
20	Qwen3 VL 8B Thinking Alibaba Cloud / Qwen Team	9B	—	—
21	DeepSeek VL2 DeepSeek	27B	—	—
22	DeepSeek VL2 Tiny DeepSeek	3B	—	—
23	Qwen3 VL 4B Thinking Alibaba Cloud / Qwen Team	4B	262K	$0.10 / $1.00
24	North Micro Vision Instruct Cohere	2B	—	—

Notice missing or incorrect data?

About this benchmark

What is OCRBench?

OCRBench: Comprehensive evaluation benchmark for assessing Optical Character Recognition (OCR) capabilities in Large Multimodal Models across text recognition, scene text VQA, and document understanding tasks

OCRBench is a multimodal benchmark evaluating models on image to text and vision tasks. LLM Stats tracks 24 models on this benchmark, scored on a 0–1 scale. The current average is 0.9, with the leader at 0.9.

Compare leaders on the best AI for image to text and best AI for vision leaderboards.

Current leaders

Kimi K2.5 from Moonshot AI currently leads the OCRBench leaderboard with a score of 0.923 across 24 evaluated AI models.

Kimi K2.5Moonshot AI92.3%

Qwen3.5-122B-A10BAlibaba Cloud / Qwen Team92.1%

Qwen3 VL 235B A22B InstructAlibaba Cloud / Qwen Team92.0%

FAQ

Common questions about the OCRBench benchmark and leaderboard.

What is the OCRBench benchmark?

What is the OCRBench leaderboard?

The OCRBench leaderboard ranks 24 AI models based on their performance on this benchmark. Currently, Kimi K2.5 by Moonshot AI leads with a score of 0.923. The average score across all models is 0.866.

What is the highest OCRBench score?

The highest OCRBench score is 0.923, achieved by Kimi K2.5 from Moonshot AI.

How many models are evaluated on OCRBench?

24 models have been evaluated on the OCRBench benchmark, with 0 verified results and 24 self-reported results.

Where can I find the OCRBench paper?

The OCRBench paper is available at https://arxiv.org/abs/2305.07895. The paper details the methodology, dataset construction, and evaluation criteria.

What categories does OCRBench cover?

OCRBench is categorized under image to text and vision. The benchmark evaluates multimodal models.

What is the best open-source model on OCRBench?

Kimi K2.5 by Moonshot AI is the top-ranked open-source model on OCRBench, with a score of 0.923 (rank #1).

Which model offers the best value on OCRBench?

Among models scoring within 10% of the leader, Qwen3 VL 4B Instruct from Alibaba Cloud / Qwen Team is the cheapest, at $0.10 per million input tokens with a score of 0.881.

How recent are the OCRBench leaderboard results?

The OCRBench leaderboard was last updated in August 2026 and currently includes 24 evaluated models.