Across 8 shared benchmarks, InternVL-3-8B scores higher on 6 and Phi 4 Multimodal Instruct on 2. The widest gap is TextVQA-val, where InternVL-3-8B scores 82.2 against 39.9.
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.