Mono3DRefer (M3Sun) - Acc@0.5: leaderboard
Monocular 3D visual grounding on the Mono3DRefer test set: a general-purpose VLM predicts the referred object's 3D bounding box from an image and language. This is the M3Sun paper's direct VLM baseline protocol; task-trained vision models and the multi-tool M3SunAgent have separate protocols. Checkpoint names and unspecified API effort are retained as printed. Accuracy uses a 3D IoU threshold of 0.5.
Metric: 3D grounding accuracy (%) at IoU 0.5. Source: arxiv.org. Saturation forecast: Around 2030. 6 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 0.23 | #185 |
| 2 | O3 | 0.18 | #242 |
| 3 | Gemini 2.5 Pro | 0.07 | #280 |
| 4 | MiMo-VL-7B | 0.05 | #1457 |
| 5 | Gemini 2.5 Flash | 0 | #466 |
| 6 | Qwen 2.5 VL 72B | 0 | #782 |
No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=mono3drefer-m3sun-acc-0-5 · How It Works · Data refreshed daily, snapshot 2026-11-22.