GST-Bench - Top-Down Selection (Medium): leaderboard
Metric: Accuracy (%; multiple choice of the top-down image of the explored scene, medium 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: Around December 2026. 22 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3 Pro | 82.73 |
| 2 | Gemini 2.5 Pro | 80 |
| 3 | GPT-5 | 79.55 |
| 4 | GPT-4o | 71.36 |
| 5 | Seed 1.8 | 67.27 |
| 6 | Qwen 3 VL 32B | 60.45 |
| 7 | InternVL3.5-8B | 35 |
Interactive version: theaggregate.ai/benchmark?slug=gst-bench-top-down-selection-medium · How It Works · Data refreshed daily, snapshot 2026-09-29.