CapArena-Auto: leaderboard
CapArena-Auto evaluates image-caption quality through pairwise battles against captioning baselines.
Metric: CapArena-Auto Score (avg). Source: huggingface.co. Saturation forecast: Estimated already saturated. 22 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 1.5 Pro (002) | 56.2 | #349 |
| 2 | GPT-4o (2024-08-06) | 44 | #336 |
| 3 | Qwen 2.5 VL 72B Instruct | 35.3 | #377 |
| 4 | Gemini 2.0 Flash (Preview) | 30.8 | #370 |
| 5 | Claude 3.5 Sonnet (20240620) | 21.5 | #299 |
| 6 | GPT-4o Mini (2024-07-18) | 9.3 | #558 |
| 7 | Llama 3.2 90B Vision Instruct | -25.7 | #622 |
| 8 | MiniCPM-V-2.6 | -38 | #836 |
| 9 | Qwen 2 VL 7B Instruct | -49 | #826 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=caparena-auto · How It Works · Data refreshed daily, snapshot 2026-10-20.