Image Captioning on CapArena-Auto
91.7Average Win-Rate MarginReflectCAP
Evaluation Results
| Method | Links | |
|---|---|---|
| ReflectCAPBase Model=Qwen3-VL-32B2026.04 | 91.7 | |
| ReflectCAPBase Model=GPT-4.1-mini2026.04 | 90 | |
| Qwen3-VL-32BStrategy=Zero-shot2026.04 | 87.3 | |
| CapMASBase Model=Qwen3-VL-32B2026.04 | 87 | |
| ReflectCAPBase Model=Qwen3-VL-8B2026.04 | 85.3 | |
| Qwen3-VL-8BStrategy=Zero-shot2026.04 | 76 | |
| CapMASBase Model=Qwen3-VL-8B2026.04 | 74 | |
| GPT-5.2Strategy=Zero-shot2026.04 | 70 | |
| Gemini-1.5-Pro†Strategy=Zero-shot2026.04 | 62.3 | |
| GPT-4.1-miniStrategy=Zero-shot2026.04 | 57.7 | |
| InternVL3.5-4BStrategy=Zero-shot2026.04 | 54.3 | |
| CapMASBase Model=GPT-4.1-mini2026.04 | 53.7 | |
| ReflectCAPBase Model=GPT-4.1-nano2026.04 | 51.3 | |
| CapMASBase Model=InternVL3.5-4B2026.04 | 46.7 | |
| GPT-4o-0806†Strategy=Zero-shot2026.04 | 44.3 | |
| Qwen2.5VL-72B†Strategy=Zero-shot2026.04 | 39.7 | |
| Qwen2.5-VL-7BStrategy=Zero-shot2026.04 | 33.7 | |
| Claude-3.5-Sonnet†Strategy=Zero-shot2026.04 | 29.7 | |
| CapMASBase Model=Qwen2.5-VL-7B2026.04 | 28 | |
| ReflectCAPBase Model=Qwen2.5-VL-32B2026.04 | 25.3 | |
| InternVL3.5-38BStrategy=Zero-shot2026.04 | 24 | |
| GPT-4.1-nanoStrategy=Zero-shot2026.04 | 21.2 | |
| ReflectCAPBase Model=InternVL3.5-4B2026.04 | 18.7 | |
| CapMASBase Model=InternVL3.5-38B2026.04 | 15 | |
| CapMASBase Model=GPT-4.1-nano2026.04 | 14.7 | |
| ReflectCAPBase Model=InternVL3.5-38B2026.04 | 12 | |
| CapMASBase Model=Qwen2.5-VL-32B2026.04 | 8 | |
| ReflectCAPBase Model=Qwen2.5-VL-7B2026.04 | 7.3 | |
| Qwen2.5-VL-32BStrategy=Zero-shot2026.04 | 5.7 |