GST-Bench - Trajectory Selection: leaderboard

Metric: Accuracy (%; multiple choice of the camera trajectory drawn on the top-down image; 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 July 2028. 22 models tracked.

Top models

#ModelScore
1Gemini 3 Pro42.18
2GPT-539.34
3Gemini 2.5 Pro37.44
4Seed 1.831.75
5Qwen 3 VL 32B29.38
6GPT-4o25.12
7InternVL3.5-8B25.12

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