DiffCap-Bench - Background: leaderboard
Metric: Macro-averaged F1* (%) on background changes: 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: Estimated already saturated. 16 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Kimi K2.5 (Non-reasoning) | 90 |
| 2 | Gemini 3 Pro | 89.4 |
| 3 | GPT-5.2 | 89 |
| 4 | Kimi K2.5 (Thinking) | 86.9 |
| 5 | Qwen 3 VL 32B (Thinking) | 85.7 |
| 6 | Step3 VL 10B | 84.1 |
| 7 | Grok 4.20 | 82.7 |
| 8 | Qwen 3 VL 8B (Thinking) | 81.4 |
| 9 | Qwen 3 VL 32B Instruct | 81.3 |
| 10 | Qwen 3 VL 8B Instruct | 77.3 |
Interactive version: theaggregate.ai/benchmark?slug=diffcap-bench-background · How It Works · Data refreshed daily, snapshot 2026-10-07.