InternVL2.5-8B: benchmark results
Provider: Shanghai AI Lab. Released 2024-12-05. Access: Open.
Aggregate Rating 1493 ± 8, rank #2471 of 10916 rated models, from 1130 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| EviPathBench - Diagnostic Reasoning | 93.44 | Accuracy (%) pooled over all questions (the All part of each | 100 |
| LabRobFail-Bench - Failure Classification | 42.06 | Top-1 accuracy (%; the failure category among five (percepti | 100 |
| MiSCHiEF (Safety) - Image-to-Caption Matching | 87.37 | Accuracy (%, 0-100) on MiS: 190 safety minimal pairs, a safe | 100 |
| SIS-Bench - Object Attribute | 82.9 | Accuracy (%; 387 object attribute questions; four-option mul | 100 |
| MEGA-Bench Task - MFC Bench Check Face Swap | 71.4 | Task Score (%) | 98.1 |
| MEGA-Bench Task - Entertainment Web Game Style | 89.3 | Task Score (%) | 97.7 |
| MEGA-Bench Task - Logical Reasoning Fit Pattern | 50 | Task Score (%) | 97.7 |
| OpenVLM MMMU - Music | 46.7 | Accuracy (%) | 96.8 |
| OpenVLM MMBench V1.1 CN - Image Scene | 92.4 | Accuracy (%) | 96.6 |
| OpenVLM MMBench V1.1 EN - Image Scene | 91.7 | Accuracy (%) | 96.6 |
| MEGA-Bench Task - Video Action Recognition | 85.7 | Task Score (%) | 96.5 |
| OpenVLM MMBench V1.1 CN - Attribute Recognition | 96.6 | Accuracy (%) | 96.5 |
Interactive version: theaggregate.ai/model?slug=internvl2-5-8b · How It Works · Data refreshed daily, snapshot 2026-11-08.