OpenGVLab: InternVL3 14B

InternVL3 14B, an advanced multimodal large language model from opengvlab, demonstrates strong overall performance with a focus on multimodal perception and reasoning. The model consistently ranks among the fastest, achieving the 89th percentile across three benchmarks, indicating efficient processing. It also offers competitive pricing, typically providing cost-effective solutions at the 66th percentile. Notably, InternVL3 14B exhibits exceptional reliability, achieving a 100% success rate across all benchmarks, signifying minimal technical failures and consistent response delivery. In terms of specific performance, the model shows varied strengths across categories. While its General Knowledge accuracy is solid at 97.5%, it falls in the 54th percentile, suggesting room for improvement in highly specialized or obscure topics. Similarly, its Ethics performance, at 98.0% accuracy, is in the 43rd percentile. However, InternVL3 14B excels in Email Classification, achieving 99.0% accuracy and ranking in the 87th percentile, highlighting a key strength in categorization tasks. A notable weakness is its relatively high duration for General Knowledge and Ethics benchmarks, ranking in the 95th and 96th percentiles respectively, indicating slower processing for these complex tasks despite its overall speed ranking. Its core strengths lie in its multimodal capabilities, exceptional reliability, and strong performance in classification tasks.