Across 10 shared benchmarks, InternVL-2-4B scores higher on 5 and Phi 3.5 Vision Instruct on 5. The widest gap is InterGPS (test), where InternVL-2-4B scores 45.6 against 36.3.
| Benchmark | InternVL-2-4B | Phi 3.5 Vision Instruct |
|---|---|---|
| AI2D | 77.3 | 78.1 |
| BLINK | 45.9 | 57 |
| InterGPS (test) | 45.6 | 36.3 |
| MathVista | 53.7 | 43.9 |
| MMBench (dev-en) | 83.4 | 81.9 |
| MMMU (val) (Pass@1) | 44.2 | 43 |
| POPE (test) | 83.3 | 86.1 |
| ScienceQA (img-test) | 94.9 | 91.3 |
| TextVQA (val) | 66.2 | 72 |
| Video-MME Overall | 49.9 | 50.8 |
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.