DVBench: leaderboard
DVBench average accuracy on the closed-ended questions about 300 real-world data videos, which combine animated charts with narration.
Metric: Average accuracy (%, closed-ended questions). Source: bomiaowang.github.io. Saturation forecast: Around February 2027. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 77.91 | #64 |
| 2 | Kimi K2.5 | 75.87 | #152 |
| 3 | Claude Sonnet 4.6 | 72.51 | #94 |
| 4 | GPT-5.4 | 65.31 | #83 |
| 5 | Qwen 3.5 27B | 62.3 | #192 |
| 6 | Qwen 3.5 397B A17B | 61.82 | #154 |
| 7 | Qwen 3.5 122B A10B | 59.3 | #184 |
| 8 | Gemma 4 31B (IT) | 58.82 | #178 |
| 9 | Qwen 3.5 9B | 52.82 | #391 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=dvbench · How It Works · Data refreshed daily, snapshot 2026-11-05.