Baidu: ERNIE 4.5 VL 424B A47B

Baidu's ERNIE 4.5 VL 424B A47B model demonstrates moderate speed performance, ranking in the 35th percentile across benchmarks, and offers competitive pricing, placing in the 44th percentile. Notably, its reliability is exceptionally strong, boasting a 94% success rate, indicating consistent and usable responses. The model exhibits strong performance in several key areas. It achieves perfect accuracy in Ethics, highlighting its robust moral reasoning capabilities, and shows high accuracy in General Knowledge (99.5%) and Email Classification (99.0%). Its Coding (89.0%) and Mathematics (88.0%) skills are also commendable. A notable strength is its ability to acknowledge uncertainty, with a 96.0% accuracy in the Hallucinations benchmark. However, the model shows a notable weakness in Reasoning, with an accuracy of 52.0%, placing it in the 35th percentile for this category. Instruction Following also presents a moderate challenge at 62.0% accuracy. Despite these areas for improvement, its multimodal MoE architecture, long-context generation (up to 131k tokens), and support for both "thinking" and non-thinking inference modes, coupled with efficient scaling capabilities, position it as a versatile tool for vision-language tasks in English and Chinese.