DiffCap-Bench - Attribute: leaderboard
Metric: Macro-averaged F1* (%) on attribute modifications such as size or colour: harmonic mean of recall* (share of key differences described) and precision* (share of described differences that are correct) on the 1,075 image pairs of DiffCap-Bench (6,713 human-verified difference items), one fixed captioning prompt, open models at temperature 0.1 and proprietary models at their defaults, a Gemini 2.5 Pro judge matching each described difference against the key difference list; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 16 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3 Pro | 80.8 |
| 2 | Kimi K2.5 (Thinking) | 80 |
| 3 | Kimi K2.5 (Non-reasoning) | 80 |
| 4 | Qwen 3 VL 32B (Thinking) | 78.6 |
| 5 | GPT-5.2 | 77.1 |
| 6 | Qwen 3 VL 8B (Thinking) | 73.3 |
| 7 | Step3 VL 10B | 70.9 |
| 8 | Qwen 3 VL 8B Instruct | 70.4 |
| 9 | Grok 4.20 | 70.3 |
| 10 | Qwen 3 VL 32B Instruct | 68.8 |
Interactive version: theaggregate.ai/benchmark?slug=diffcap-bench-attribute · How It Works · Data refreshed daily, snapshot 2026-10-07.