Technology ❯ Artificial Intelligence ❯ Model Performance

Benchmarking

Comparative Analysis Multimodal Models MMLU Scores International Mathematics Olympiad MMLU Benchmark Vision-Language Tasks Orion Chatbot Arena Coding Tests DeepSeek Reinforcement Learning Gemini 2.0 Pro User Experience SWE-Bench Scores Llama-4 Models Latency-Sensitive Tasks Coding Efficiency Coding Proficiency Real-world Applications Gemini 2.5 Flash-Lite Performance Metrics USAMO 2025 User Reviews o3-pro Deception Metrics Reasoning Tasks GPT Models Vision Workloads OmniBenchDoc MMLU and MGSM Scores Multilingual MMLU

PaddlePaddle Releases PaddleOCR-VL, a 0.9B Vision-Language Model for Multilingual Document Parsing

The 0.9B model pairs a NaViT-style visual encoder with an ERNIE-4.5-0.3B language model for multilingual document parsing across 109 languages, with SOTA results reported by the authors.