E-VAds: leaderboard
Metric: Strict score (%): share of answers the judge rates fully correct (1.0) over all five tasks; 16,384 test questions on 3,389 Taobao e-commerce short videos, video frames plus the ASR transcript; Qwen3-Coder-Plus judge grading each open-ended answer against the expert-reviewed reference and its visual, ASR and OCR evidence on a five-level scale (0, 0.25, 0.5, 0.75, 1); higher is better. Source: arxiv.org. Saturation forecast: Around September 2027. 17 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Flash | 35 | #93 |
| 2 | GPT-5.2 | 27.8 | #105 |
| 3 | Qwen3 Omni 30B A3B Instruct | 20.5 | #362 |
| 4 | Qwen 3 VL 30B A3B Instruct | 19.6 | #365 |
| 5 | Qwen 3 VL 8B (Thinking) | 18.6 | |
| 6 | Qwen 3 VL 8B Instruct | 15.3 | #401 |
| 7 | Qwen 3 VL 30B A3B (Thinking) | 14.8 | #338 (Qwen 3 VL 30B A3B) |
| 8 | Qwen 2.5 VL 32B Instruct | 14.2 | #443 |
| 9 | InternVL3-8B | 12.6 | #606 |
| 10 | Qwen 2.5 VL 7B Instruct | 11.1 | #643 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=e-vads · How It Works · Data refreshed daily, snapshot 2026-10-11.