EgoPoint-Bench - Simulation: leaderboard

EgoPoint-Bench (2026) asks multimodal models about the object a first-person hand points at, on held-out simulation scenes and real smart-glasses recordings, zero-shot; true/false and single-choice answers by exact match and open answers scored by a GPT-4o judge. This board covers the 1,929 simulation test questions across five capability dimensions. Accuracy in percent; higher is better.

Metric: Accuracy (%) over the 1,929 held-out simulation test questions (HM3D, HSSD and AI2-THOR scenes), all five dimensions; zero-shot (direct inference); true/false and single-choice questions by exact match, open-ended answers scored 0 to 1 by a GPT-4o judge; greedy decoding for open models, temperature 0 for API models; higher is better. Source: arxiv.org. Saturation forecast: Around 2030. 12 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Qwen 3 VL 8B Instruct58.29#1029
2Qwen 3 VL 32B Instruct58.28#606
3GPT-5 Mini57.66#364
4Gemini 3 Flash57.21#175
5Gemini 3 Pro56.44#145
6GPT-5.2 Instant54.8#304
7GLM-4.6V-Flash53.29#1040
8InternVL3.5-8B-Instruct52.62#1891
9InternVL3.5-2B-Instruct51.74#6517
10InternVL3.5-14B-Instruct49.99#1736
11LLaVA-v1.5-7B48.82#7630
12LLaVA-NeXT-7B48.17#5032

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=egopoint-bench-simulation · How It Works · Data refreshed daily, snapshot 2026-11-21.