CaST-Bench (Full Set): leaderboard

Multiple-choice answer accuracy in percent on the full CaST-Bench set (2026) of 2,066 six-option video questions over 1,015 videos, where the model first outputs a grounded spatio-temporal causal chain and then answers; chance is 16.7 percent and higher is better.

Metric: Answer accuracy (%) on the multiple-choice questions (chance 16.67), all 2,066 six-option questions over 1,015 videos; every model must first output a grounded spatio-temporal causal chain (evidence segments with per-second boxes) and then answer; native video input except the GPT-5 models, which get 1 fps frames; higher is better. Source: arxiv.org. Saturation forecast: Around 2029. 15 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Gemini 2.5 Pro50.34#276
2GPT-546.32#178
3Gemini 2.5 Flash45.6#463
4Qwen 3 VL 4B Instruct45.3#1332
5InternVL3.5-30B-A3B44.53#1267
6InternVL3.5-14B43.27#1504
7Qwen 3 VL 8B Instruct43.13#1029
8Qwen 2.5 VL 7B Instruct41.09#1902
9InternVL3.5-8B40.95#1604
10GLM-4.1V-9B (Thinking)39.55
11InternVL2.5-8B39.55#2370
12Qwen 3 VL 8B (Thinking)39.16#910
13GPT-5 Mini37.22#364
14MiMo-VL-7B-RL-250832.24#823
15LLaVA-NeXT-Video-34B28.17#4307

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

Interactive version: theaggregate.ai/benchmark?slug=cast-bench-full-set · How It Works · Data refreshed daily, snapshot 2026-11-21.