Across 10 shared benchmarks, Gemini 1.5 Pro scores higher on 8 and InternVL-2-4B on 2. The widest gap is BLINK, where Gemini 1.5 Pro scores 61 against 45.9.
| Benchmark | Gemini 1.5 Pro | InternVL-2-4B |
|---|---|---|
| AI2D | 80.9 | 77.3 |
| BLINK | 61 | 45.9 |
| InterGPS (test) | 58.2 | 45.6 |
| MathVista | 70.6 | 53.7 |
| MMBench (dev-en) | 87.9 | 83.4 |
| MMMU (val) (Pass@1) | 54.1 | 44.2 |
| POPE (test) | 89.3 | 83.3 |
| ScienceQA (img-test) | 86 | 94.9 |
| TextVQA (val) | 64.5 | 66.2 |
| Video-MME Overall | 62.6 | 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.