VGenST-Bench - Visibility Identification: 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. Visibility Identification task: decide whether one agent can see another as it moves around an occluder (vista scale, exocentric, dynamic). Accuracy in percent on multiple-choice questions under circular evaluation (right under every rotation of the options), 8 sampled frames; higher is better.

Metric: Visibility Identification (vista scale, exocentric, dynamic) 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
1GPT-5.480.6#138
2Gemma 4 31B (IT)71.1#354
3Gemini 3.1 Flash Lite71.1#377
4Gemini 3 Flash66.1#180
5Gemma 4 26B A4B (IT)66#563
6Qwen 3.5 27B64.7#380
7InternVL3.5-30B-A3B63.5#1269
8Qwen 3.5 9B62.1#838
9GPT-5.4 Mini60.2#409
10Qwen 3.5 4B59.5#1270
11Kimi K2.656.4#197
12GLM-4.6V-Flash52.8#1049
13InternVL3.5-8B49.2#1618
14GPT-5.4 Nano48.8#653
15InternVL3.5-4B42#1946

No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=vgenst-bench-visibility-identification · How It Works · Data refreshed daily, snapshot 2026-11-22.