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
| # | Model | Score |
|---|---|---|
| 1 | Seed 1.8 | 99.07 |
| 2 | Gemini 3 Pro | 98.15 |
| 3 | Gemini 2.5 Pro | 97.69 |
| 4 | GPT-5 | 97.69 |
| 5 | Qwen 3 VL 32B | 94.44 |
| 6 | GPT-4o | 81.02 |
| 7 | InternVL3.5-8B | 75 |
Interactive version: theaggregate.ai/benchmark?slug=gst-bench-top-down-selection-easy · How It Works · Data refreshed daily, snapshot 2026-09-29.