GST-Bench - Egocentric Direction (Semantic): leaderboard
GST-Bench (2026) asks vision-language models about long egocentric videos rendered in a simulator, where answers require a global map of the scene rather than the current frame. The model gives the direction from its current viewpoint to an object it saw earlier, named by its category; mean accuracy within 15, 30 and 45 degrees, in percent; higher is better.
Metric: Mean accuracy (%; share of predicted egocentric directions to a previously seen object within 15, 30 and 45 degrees, averaged; target named by its category; 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 September 2027. 22 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 31.09 | #167 |
| 2 | Gemini 2.5 Pro | 23.08 | #266 |
| 3 | Gemini 3 Pro | 22.11 | #137 |
| 4 | Seed 1.8 | 20.35 | #242 |
| 5 | GPT-4o | 13.46 | #741 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=gst-bench-egocentric-direction-semantic · How It Works · Data refreshed daily, snapshot 2026-11-13.