PinpointQA - Structured Spatial Prediction: leaderboard
Structured spatial prediction on PinpointQA (2026): from 64 frames of an indoor ScanNet++ or ScanNet200 video, the model outputs the supporting surface, reference objects, relations and distances of a small object as JSON, scored programmatically and zero when no JSON is recoverable. Mean score scaled to 0-100; higher is better.
Metric: Structured spatial prediction score (times 100, micro over QA pairs): programmatic weighted score of supporting surface, reference objects, relations and distances in a recovered JSON prediction (zero when no JSON is recoverable) of the PinpointQA evaluation split (2,019 QA pairs from held-out ScanNet++ and ScanNet200 scenes; 64 uniformly sampled RGB frames per indoor video, no depth or 3D input, greedy decoding); higher is better. Source: arxiv.org. Saturation forecast: Around 2029. 8 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.4 | 15 | #135 |
| 2 | Kimi K2.5 | 15 | #266 |
| 3 | Qwen 3 VL 8B Instruct | 12 | #1029 |
| 4 | SenseNova-SI-1.3-InternVL3-8B | 12 | #1610 |
| 5 | InternVL3.5-8B-Instruct | 9 | #1891 |
| 6 | LLaVA-OneVision-1.5-8B | 7 | #1864 |
| 7 | Cambrian-S-7B | 5 | #2198 |
| 8 | Spatial-MLLM-820K-4B | 0 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=pinpointqa-structured-spatial-prediction · How It Works · Data refreshed daily, snapshot 2026-11-21.