PinpointQA - Nearest Reference Identification: leaderboard

Nearest reference identification on PinpointQA (2026): from 64 frames of an indoor ScanNet++ or ScanNet200 video, the model picks which of four objects is nearest to a small target, not counting its supporting surface. Exact-match accuracy scaled to 0-100; higher is better.

Metric: Exact-match accuracy (times 100, micro over QA pairs) of four-option nearest reference object identification (excluding the supporting surface) 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 July 2028. 8 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Kimi K2.542#272
2GPT-5.439#138
3Qwen 3 VL 8B Instruct37#1043
4InternVL3.5-8B-Instruct36#1911
5SenseNova-SI-1.3-InternVL3-8B36#1627
6Cambrian-S-7B33#2216
7LLaVA-OneVision-1.5-8B30#1876
8Spatial-MLLM-820K-4B30

No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=pinpointqa-nearest-reference-identification · How It Works · Data refreshed daily, snapshot 2026-11-22.