CiteVQA - Evidence Relevance: leaderboard
Metric: Relevance of each cited region to its answer, judged on a 0-5 scale and multiplied by 20 (0-100), over all 1,897 questions, the model answers from the document pages and cites element-level bounding boxes, unified prompt, temperature 1.0, at most 4,096 output tokens, Qwen3-VL-235B-A22B judge; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 20 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) (High) | 83.6 |
| 2 | Gemini 3 Flash (Preview) (High) | 75.7 |
| 3 | GPT-5.4 (xHigh) | 67.5 |
| 4 | GPT-5.2 (xHigh) | 56.6 |
| 5 | Seed 2.0 Pro | 54.9 |
| 6 | Qwen 3 VL 235B A22B Instruct | 35.3 |
| 7 | Gemma 4 31B | 35 |
| 8 | Qwen 3 VL 32B Instruct | 30.5 |
| 9 | GLM-5V Turbo | 29.2 |
| 10 | Kimi K2.5 | 26.8 |
| 11 | Qwen 3.5 27B | 25.3 |
| 12 | Qwen 3.6 Plus | 25 |
| 13 | Qwen 3.5 397B A17B | 24.6 |
| 14 | Qwen 3.5 122B A10B | 19 |
| 15 | Gemma 4 26B A4B | 17.9 |
Interactive version: theaggregate.ai/benchmark?slug=citevqa-evidence-relevance · How It Works · Data refreshed daily, snapshot 2026-10-07.