DashBench - Q1: leaderboard

DashBench evaluates video understanding on 1,709 driving clips. Five explanation prompts and binary event detection are scored separately; mean agreement excludes the binary task. The 823-clip corner-case-window subset and training-free decoder variants retain their own conditions. Q5 asks for definition-guided corner-case identification and description. This board reports lLM-judged answer accuracy (%). Higher is better.

Metric: LLM-judged answer accuracy (%). Source: arxiv.org. Saturation forecast: Around April 2027. 22 models tracked. Updated 14 Nov 2026.

Top models

#ModelSettingScoreOverall rank
1Gemini 3 Flash32.36#163
2GPT-519.95#167
3Ovis2.5-9B19.68#1236
4InternVL3.5-241B19.31#625
5Qwen 3 VL 8B18.9#899
6Qwen 3 VL 235B A22B18.72#484
7Qwen 3 VL 32B17.2#609
8MiniCPM-V-4.513.34#1390
9Tarsier2-7B13.28#844
10InternVL3.5-38B12.64#986
11GLM-4.1V-9B (Thinking)7.85
12VITA-1.57.31#3196
13InternLM-XComposer2.56.26#3981
14LlaVA-OV-RS (7B)6.09
15InternVL3.5-8BEvaluation run: DashBench TRiCD5.03#1558

No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.

Interactive version: theaggregate.ai/benchmark?slug=dashbench-q1 · How It Works · Data refreshed daily, snapshot 2026-11-16.