VGenST-Bench - Landmark Spatial Composition: leaderboard

VGenST-Bench (2026) tests spatio-temporal reasoning of multimodal LLMs on short videos synthesized from validated scene graphs with text-to-image and image-to-video models. Landmark Spatial Composition task: compose two camera moves to give the compass direction of one landmark from another (environmental scale, exocentric, static). Accuracy in percent on multiple-choice questions under circular evaluation (right under every rotation of the options), 8 sampled frames; higher is better.

Metric: Landmark Spatial Composition (environmental scale, exocentric, static) accuracy (%) on the base multiple-choice questions of VGenST-Bench under circular evaluation (correct only if right under every cyclic permutation of the options), macro-averaged over the task's QA types, 8 uniformly sampled frames from about 100 synthesized videos per task (generated with text-to-image and image-to-video models from validated scene graphs); higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 15 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Gemini 3 Flash66#175
2GPT-5.457#135
3Kimi K2.649.8#192
4Gemini 3.1 Flash Lite47.9#371
5Gemma 4 31B (IT)46.2#350
6InternVL3.5-30B-A3B45.7#1267
7Qwen 3.5 27B41.2#375
8Gemma 4 26B A4B (IT)41.1#556
9GPT-5.4 Mini41#406
10GLM-4.6V-Flash40.1#1040
11Qwen 3.5 9B38#828
12GPT-5.4 Nano37.1#651
13Qwen 3.5 4B35.6#1241
14InternVL3.5-4B32.7#1930
15InternVL3.5-8B32.6#1604

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=vgenst-bench-landmark-spatial-composition · How It Works · Data refreshed daily, snapshot 2026-11-21.