Across 10 shared benchmarks, InternVL-2-4B scores higher on 2 and InternVL-2-8B on 8. The widest gap is InterGPS (test), where InternVL-2-8B scores 53.2 against 45.6.
| Benchmark | InternVL-2-4B | InternVL-2-8B |
|---|---|---|
| AI2D | 77.3 | 81.4 |
| BLINK | 45.9 | 45.4 |
| InterGPS (test) | 45.6 | 53.2 |
| MathVista | 53.7 | 51.1 |
| MMBench (dev-en) | 83.4 | 87 |
| MMMU (val) (Pass@1) | 44.2 | 46.3 |
| POPE (test) | 83.3 | 84.2 |
| ScienceQA (img-test) | 94.9 | 95.9 |
| TextVQA (val) | 66.2 | 68.8 |
| Video-MME Overall | 49.9 | 52.6 |
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.