InternVL2.5-2B — benchmark results
Provider: Shanghai AI Lab. Released 2024-12-06. Access: Open.
Unified ELO 1275 ± 10, rank #1581 of 1776 rated models, from 780 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| MEGA-Bench Task - Bongard Problem | 81.6 | Task Score (%) | 100 |
| MEGA-Bench Task - Paper Review Rating | 76.4 | Task Score (%) | 98.8 |
| Open LMM Reasoning - WeMath - RoteMemorization (Strict) | 81.9 | Accuracy (%) | 95.1 |
| Open LMM Reasoning - MathVision - topology | 34.8 | Accuracy (%) | 94.9 |
| OpenVLM MMBench V1.1 EN - Social Relation | 91.2 | Accuracy (%) | 94.7 |
| MEGA-Bench Task - CVBench Adapted CVBench Count | 71.4 | Task Score (%) | 94.2 |
| OpenVLM MMBench V1.1 EN - Attribute Recognition | 96.6 | Accuracy (%) | 93.6 |
| MEGA-Bench Task - Relative Reflectance Of Different Regions | 50 | Task Score (%) | 93 |
| MEGA-Bench Task - NExT-QA MC | 94.7 | Task Score (%) | 91.9 |
| OpenVLM MMBench V1.1 CN - Image Scene | 91 | Accuracy (%) | 91.2 |
| MEGA-Bench Task - Symbolic Graphics Programs Scalable Vector Graphics | 16.7 | Task Score (%) | 88.4 |
| Open LMM Reasoning - WeMath - InsufficientKnowledge (Loose) | 63.8 | Accuracy (%) | 87.7 |
Interactive version: theaggregate.ai/model?slug=internvl2-5-2b · How the rankings work · Data refreshed daily, snapshot 2026-07-22.