VGenST-Bench - Multi-Container Attribute Mapping: 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. Multi-Container Attribute Mapping task: map each object to the container that holds it, seen from a first-person view (figural scale, egocentric, 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: Multi-Container Attribute Mapping (figural scale, egocentric, 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 Flash97#180
2GPT-5.488.8#138
3Gemini 3.1 Flash Lite87.9#377
4Qwen 3.5 27B87.6#380
5InternVL3.5-30B-A3B86.4#1269
6Qwen 3.5 9B85.2#838
7Gemma 4 31B (IT)82#354
8Qwen 3.5 4B81.4#1270
9Gemma 4 26B A4B (IT)77#563
10InternVL3.5-8B72.6#1618
11Kimi K2.671.2#197
12GPT-5.4 Mini68.5#409
13GLM-4.6V-Flash68#1049
14GPT-5.4 Nano64.2#653
15InternVL3.5-4B62.9#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-multi-container-attribute-mapping · How It Works · Data refreshed daily, snapshot 2026-11-22.