SceneFunRI: leaderboard
In real-world scenes, target objects may reside in regions that are not visible. While humans can often infer the locations of occluded objects from context and commonsense knowledge, this capability remains a major challenge for vision-language models (VLMs). To address this gap, we introduce SceneFunRI, a benchmark for Reasoning the Invisible. Based on the SceneFun3D dataset, SceneFunRI formulat
Metric: cacc_75 (self-reported). Source: benchmarklist.com. Saturation forecast: Around June 2027. 17 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Flash | 15.2 | #99 |
| 2 | Qwen 3.5 122B A10B | 12.63 | #173 |
| 3 | Qwen 3.5 9B | 11.11 | #379 |
| 4 | Qwen 3.5 4B | 10.99 | #475 |
| 5 | Qwen 3.5 27B | 7.95 | #187 |
| 6 | Gemma 3 27B (IT) | 7.6 | #523 |
| 7 | Gemma 3 4B (IT) | 4.21 | #992 |
| 8 | Gemma 3 12B (IT) | 4.09 | #664 |
| 9 | InternVL3-8B | 4.09 | #620 |
| 10 | InternVL3-38B | 3.86 | #405 |
| 11 | Qwen 3.5 0.8B | 3.63 | #1424 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=scenefunri · How It Works · Data refreshed daily, snapshot 2026-10-20.