MVPBench - Multi-View Perception Pairing: leaderboard
Metric: MVPBench normalized proficiency score (accuracy minus chance, divided by one minus chance, times 100, so 0 is chance level and 100 is perfect) on its 60 multi-video questions matching a first-person reference clip to one of three second-person candidates; chance is 1/3; answers extracted by rules and GPT-4-turbo; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 15 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 2.5 VL 72B Instruct | 67.5 | #364 |
| 2 | GPT-4o | 65.01 | #333 |
| 3 | Gemini 2.5 Flash | 58.5 | #237 |
| 4 | Qwen 2.5 VL 7B Instruct | 55 | #643 |
| 5 | InternVL3-78B | 35.01 | #345 |
| 6 | InternVL3-38B | 20.01 | #395 |
| 7 | InternVL3-8B | 15 | #606 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=mvpbench-multi-view-perception-pairing · How It Works · Data refreshed daily, snapshot 2026-10-11.