DashBench - Q3: 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 February 2027. 22 models tracked. Updated 14 Nov 2026.

Top models

#ModelSettingScoreOverall rank
1GPT-531.19#167
2Gemini 3 Flash29.84#163
3Qwen 3 VL 235B A22B16.79#484
4Qwen 3 VL 8B14.45#899
5Qwen 3 VL 32B12.87#609
6Tarsier2-7B11.7#844
7LlaVA-OV-RS (7B)11.59
8InternVL3.5-241B11.47#625
9InternVL3.5-38B8.66#986
10Ovis2.5-9B7.77#1236
11InternVL3.5-8BEvaluation run: DashBench TRiCD7.2#1558
12MiniCPM-V-4.56.38#1390
13InternLM-XComposer2.54.92#3981
14InternVideo2.5-Chat-8B4.15#2017
15VITA-1.53.16#3196

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

Interactive version: theaggregate.ai/benchmark?slug=dashbench-q3 · How It Works · Data refreshed daily, snapshot 2026-11-16.