GST-Bench - Top-Down Position: leaderboard
Metric: Mean accuracy (%; predicted position of the camera on the global top-down image within 100 to 300 pixels, averaged over five thresholds; 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 2029. 22 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3 Pro | 37.47 |
| 2 | Gemini 2.5 Pro | 30.47 |
| 3 | Seed 1.8 | 28.45 |
| 4 | GPT-4o | 24.03 |
| 5 | Qwen 3 VL 32B | 24.03 |
| 6 | GPT-5 | 22.33 |
| 7 | InternVL3.5-8B | 18.84 |
Interactive version: theaggregate.ai/benchmark?slug=gst-bench-top-down-position · How It Works · Data refreshed daily, snapshot 2026-09-29.