OpenGVLab: InternVL3 78B

OpenGVLab's InternVL3 78B, created on September 15, 2025, is an advanced multimodal large language model (MLLM) with a context length of 32768. It consistently ranks among the fastest models and offers highly competitive pricing across all benchmarks. Demonstrating exceptional reliability, the model achieved a 100% success rate across seven benchmarks, indicating minimal technical failures. In terms of performance, InternVL3 78B exhibits perfect accuracy in both General Knowledge and Ethics benchmarks, often being the most accurate model at its price point and speed. It also shows strong performance in Email Classification (97.0% accuracy) and Coding (84.0% accuracy). While its Instruction Following (57.6% accuracy) and Reasoning (52.0% accuracy) capabilities are moderate, they still place it within a competitive range. A notable strength is its 0.0% hallucination rate, indicating a robust ability to acknowledge uncertainty. Overall, InternVL3 78B stands out for its high accuracy in knowledge-based and ethical tasks, combined with superior speed, cost-efficiency, and reliability.