Across 10 shared benchmarks, GPT-4o mini scores higher on 7 and InternVL-2-4B on 3. The widest gap is Video-MME Overall, where GPT-4o mini scores 61.2 against 49.9.
| Benchmark | GPT-4o mini | InternVL-2-4B |
|---|---|---|
| AI2D | 75.2 | 77.3 |
| BLINK | 51.9 | 45.9 |
| InterGPS (test) | 39.9 | 45.6 |
| MathVista | 56.7 | 53.7 |
| MMBench (dev-en) | 83.8 | 83.4 |
| MMMU (val) (Pass@1) | 52.1 | 44.2 |
| POPE (test) | 83.6 | 83.3 |
| ScienceQA (img-test) | 84 | 94.9 |
| TextVQA (val) | 70.9 | 66.2 |
| Video-MME Overall | 61.2 | 49.9 |
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.