VGenST-Bench - Container Intersection Inference: 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. Container Intersection Inference task: decide which objects appear in both of two side-by-side containers (figural 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: Container Intersection Inference (figural 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: Estimated already saturated. 15 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Gemini 3.1 Flash Lite97#371
2Gemini 3 Flash95.7#175
3GPT-5.492.5#135
4Gemma 4 31B (IT)89.4#350
5Kimi K2.689.1#192
6Gemma 4 26B A4B (IT)84.4#556
7Qwen 3.5 27B77.9#375
8GLM-4.6V-Flash76.9#1040
9Qwen 3.5 9B75.8#828
10GPT-5.4 Mini75.4#406
11Qwen 3.5 4B72.9#1241
12InternVL3.5-30B-A3B71.6#1267
13GPT-5.4 Nano68.2#651
14InternVL3.5-8B57.6#1604
15InternVL3.5-4B55.2#1930

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-container-intersection-inference · How It Works · Data refreshed daily, snapshot 2026-11-21.