EgoPoint-Bench - Simulation Adversarial Resilience: 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 Adversarial Resilience questions (counterfactual, fallacious or void references) among the 1,929 simulation test questions. Accuracy in percent; higher is better.

Metric: Accuracy (%) on the Adversarial Resilience questions among the 1,929 held-out simulation test questions (HM3D, HSSD and AI2-THOR scenes); 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 2029. 12 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Qwen 3 VL 32B Instruct60.23#606
2Gemini 3 Flash58.39#175
3Gemini 3 Pro57.16#145
4Qwen 3 VL 8B Instruct52.11#1029
5GPT-5.2 Instant50.88#304
6LLaVA-NeXT-7B46.59#5032
7LLaVA-v1.5-7B41.92#7630
8GLM-4.6V-Flash36.17#1040
9GPT-5 Mini36.09#364
10InternVL3.5-8B-Instruct35.63#1891
11InternVL3.5-14B-Instruct33.56#1736
12InternVL3.5-2B-Instruct26.05#6517

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-adversarial-resilience · How It Works · Data refreshed daily, snapshot 2026-11-21.