DashBench - Q5: 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-539.15#167
2Gemini 3 Flash34.99#163
3Qwen 3 VL 235B A22B21.71#484
4Qwen 3 VL 8B21.53#899
5Ovis2.5-9B20.55#1236
6Qwen 3 VL 32B18.31#609
7InternVL3.5-241B17.61#625
8InternVL3.5-8BEvaluation run: DashBench TRiCD14.69#1558
9MiniCPM-V-4.513.53#1390
10InternVL3.5-38B12.93#986
11Tarsier2-7B12.35#844
12InternVideo2.5-Chat-8B10.12#2017
13GLM-4.1V-9B (Thinking)9.91
14InternLM-XComposer2.58.54#3981
15LlaVA-OV-RS (7B)7.49

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

Interactive version: theaggregate.ai/benchmark?slug=dashbench-q5 · How It Works · Data refreshed daily, snapshot 2026-11-16.