BenchLMM: leaderboard
BenchLMM — cross-style robustness benchmark for large multimodal models, testing photographic, artistic and sensor-style image distribution shifts.
Metric: GPT-3.5 score. Source: huggingface.co. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4V | 58.37 | |
| 2 | Sphinx-V2-1K | 57.43 | |
| 3 | LLaVA-1.5-13B | 55.53 | |
| 4 | LLaVA-1.5-7B | 46.83 | |
| 5 | InstructBLIP-13B | 45.03 | |
| 6 | InstructBLIP-7B | 44.63 | |
| 7 | LLaVA-1-13B | 43.5 | |
| 8 | Otter-7B | 39.13 | |
| 9 | MiniGPT4-13B | 34.93 | |
| 10 | MiniGPTv2-7B | 30.1 |
Interactive version: theaggregate.ai/benchmark?slug=benchlmm · How It Works · Data refreshed daily, snapshot 2026-10-20.