ocrbench.

OCRBench overall score on the native 0-1000 scale (card reports 92.3 normalized); max 64k tokens; avg@3; Thinking mode.; PWC evaluation id 1252; paper: Kimi K2.5: Visual Agentic Intelligence

Table 2 of Qwen3-VL technical report (arXiv:2511.21631), OCRBench (rescaled from 87.5/92.0 to the 0-1000 scale used by the existing rows).; PWC evaluation id 4749; paper: Qwen3-VL Technical Report

PWC OCRBench Score normalized to the 0-1000 convention.; Paper Table 3; SenseNova-U1-A3B-MoT Think mode on OCRBench.; PWC evaluation id 5612; paper: SenseNova-U1: Unifying Multimodal Understanding and Generation with NEO-unify Architecture

PWC OCRBench Score normalized to the 0-1000 convention.; Paper Table 6, Vision->Text, OCRBench document understanding benchmark.; PWC evaluation id 5329; paper: Qwen3.5-Omni Technical Report

PWC evaluation id 768; paper: InternVL3: Exploring Advanced Training and Test-Time Recipes for Open-Source Multimodal Models

PWC OCRBench Score normalized to the 0-1000 convention.; OCRBench row from Qwen3.6 model-card Vision benchmark table; imported as configured OCRBench Score. Source: Qwen3.6-27B Hugging Face model card benchmark table (https://huggingface.co/Qwen/Qwen3.6-27B).; PWC evaluation id 5562; paper: Qwen3.6

Reported on OCRBench (raw 0-1000 score) by Qwen3-VL-8B-Instruct model card.; PWC evaluation id 1279; paper: Qwen3-VL Technical Report

PWC OCRBench Score normalized to the 0-1000 convention.; OCRBench row from Qwen3.6 model-card Vision benchmark table; imported as configured OCRBench Score. Source: Qwen3.6-27B Hugging Face model card benchmark table (https://huggingface.co/Qwen/Qwen3.6-27B).; PWC evaluation id 5563; paper: Qwen3.6

Table 5, OCRBench. Source: Qwen2.5-VL Technical Report (arXiv:2502.13923). Model: Qwen2.5-VL-72B.; PWC evaluation id 5023; paper: Qwen2.5-VL Technical Report

OCRBench standard Score (0-1000); 880.; PWC evaluation id 1197; paper: Qianfan-OCR: A Unified End-to-End Model for Document Intelligence

PWC OCRBench Score normalized to the 0-1000 convention.; Table 3, OpenCompass suite; OCRBench (OCR). 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 5583; paper: Ovis2.5 Technical Report

PWC evaluation id 143; paper: Qwen2-VL: Enhancing Vision-Language Model's Perception of the World at Any Resolution

Instruct mode from the openbmb/MiniCPM-o-4_5 Hugging Face model card (https://huggingface.co/openbmb/MiniCPM-o-4_5); 9B params; results reported in instruct mode/variant; from the 'Image Understanding (Instruct)' table; metric label in card: OCRBench; 0-1000 scale used by other rows on this leaderboard.; PWC evaluation id 1171; paper: MiniCPM-o 4.5: Towards Real-Time Full-Duplex Omni-Modal Interaction

Table 2 of Qwen3-VL technical report (arXiv:2511.21631), OCRBench (rescaled from 87.5/92.0 to the 0-1000 scale used by the existing rows).; PWC evaluation id 4748; paper: Qwen3-VL Technical Report

Kimi-VL-A3B-Thinking-2506 on OCRBench overall score (raw 0-1000 scale) from the moonshotai/Kimi-VL-A3B-Thinking-2506 HF model card.; PWC evaluation id 3371; paper: Kimi-VL Technical Report

Kimi-VL-A3B-Instruct on OCRBench overall score (raw 0-1000 scale) from Kimi-VL Technical Report Table 3 and the moonshotai/Kimi-VL-A3B-Instruct HF model card.; PWC evaluation id 3351; paper: Kimi-VL Technical Report

PWC evaluation id 144; paper: Qwen2-VL: Enhancing Vision-Language Model's Perception of the World at Any Resolution

PWC evaluation id 886; paper: MiniMax-01: Scaling Foundation Models with Lightning Attention

paper table; source label OCRBench; metric reported as Score. Imported while expanding from ScreenSpot-Pro source papers.; PWC evaluation id 5377; paper: Qwen2.5-VL Technical Report

OCRBench full benchmark score from the Infinity-Parser2-Pro Hugging Face card / GitHub performance table. Source reports 86.20 on a 0-100 scale; stored as 862.0 on the 0-1000 OCRBench Score convention used by existing rows.; PWC evaluation id 4966; paper: Infinity-Parser2-Pro

OCRBench overall score from the dots.mocr Hugging Face model card (section 3, General Vision Tasks). The card reports 86.0 on a 0-100 scale; converted to 860 on the standard 0-1000 OCRBench scale used by other rows on this leaderboard (consistent with Qwen3-VL-2B = 85.8 -> 858 in the same table).; PWC evaluation id 1153; paper: Multimodal OCR: Parse Anything from Documents

PWC evaluation id 957; paper: HunyuanOCR Technical Report

Thinking mode from the MiniCPM-V 4.6 Hugging Face model card; official checkpoint; visual token compression ratio 16x; metric label in card: OCRBench.; PWC evaluation id 1115; paper: A Pocket-Sized MLLM for Ultra-Efficient Image and Video Understanding on Your Phone

DAMO-NLP-SG/VideoLLaMA3-7B-Image checkpoint; numbers from the 7B-Image model card main-results table.; PWC evaluation id 1214; paper: VideoLLaMA 3: Frontier Multimodal Foundation Models for Image and Video Understanding

PWC evaluation id 145; paper: Qwen2-VL: Enhancing Vision-Language Model's Perception of the World at Any Resolution

OCRBench overall score (0-1000 scale; card reports 79.8 normalised). Reported in the ZAYA1-VL-8B technical report (Zyphra). Evaluated on the Zyphra eval harness based on VLMEvalKit.; PWC evaluation id 1230; paper: ZAYA1-VL-8B Technical Report

paper table; source label OCRBench; metric reported as Score. Imported while expanding from ScreenSpot-Pro source papers.; PWC evaluation id 5378; paper: Qwen2.5-VL Technical Report

PWC evaluation id 115; paper: VideoLLaMA 3: Frontier Multimodal Foundation Models for Image and Video Understanding

Paper Table 5 OCR benchmark result for MiniCPM-Llama3-V 2.5; source reports OCRBench score.; PWC evaluation id 5183; paper: MiniCPM-V: A GPT-4V Level MLLM on Your Phone