GST-Bench - Top-Down Selection (Easy): leaderboard

Metric: Accuracy (%; multiple choice of the top-down image of the explored scene, easy distractors; zero-shot, greedy decoding, official prompt templates, over human-verified questions on 6,790 minutes of simulator-rendered egocentric exploration video). Source: arxiv.org. Saturation forecast: Estimated already saturated. 22 models tracked.

Top models

#ModelScore
1Seed 1.899.07
2Gemini 3 Pro98.15
3Gemini 2.5 Pro97.69
4GPT-597.69
5Qwen 3 VL 32B94.44
6GPT-4o81.02
7InternVL3.5-8B75

Interactive version: theaggregate.ai/benchmark?slug=gst-bench-top-down-selection-easy · How It Works · Data refreshed daily, snapshot 2026-09-29.