Across 6 shared benchmarks, InternVL3.5-8B-Instruct scores higher on 0 and Qwen3.5 27B on 6. The widest gap is ERQA, where Qwen3.5 27B scores 60.5 against 42.
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.