NoteVQA: leaderboard
Real questions photographed by hobbyist communities - pets, exotics, plants, health, fashion, jewellery, appliances, food - answered from the image. Overall averages the twelve categories; higher is better. Source: arxiv.org/abs/2609.15695.
Metric: Accuracy (%, thinking mode). Source: arxiv.org. Saturation forecast: Around April 2028. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) (Thinking) | 52.8 | #64 (Gemini 3.1 Pro (Preview)) |
| 2 | Kimi K2.6 (Thinking) | 47.6 | #113 (Kimi K2.6) |
| 3 | Claude Opus 4.7 (Thinking) | 43.7 | #51 (Claude Opus 4.7) |
| 4 | Kimi K3 (Thinking) | 43.3 | #35 (Kimi K3) |
| 5 | Qwen 3.5 397B A17B (Thinking) | 35.7 | #154 (Qwen 3.5 397B A17B) |
| 6 | Claude Sonnet 4.6 (Thinking) | 35.3 | #94 (Claude Sonnet 4.6) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=notevqa · How It Works · Data refreshed daily, snapshot 2026-11-05.