Across 8 shared benchmarks, InternVL-3-8B scores higher on 6 and Qwen2.5 VL 3B on 2. The widest gap is OCRVQA_TEST, where Qwen2.5 VL 3B scores 69.2 against 39.
Best tracked score per model per benchmark (default configuration; source-attributed). ↓ marks lower-is-better metrics. Open either model for its full surface, provenance and pricing.