DiffCap-Bench - Camera and Viewpoint: leaderboard

DiffCap-Bench (2026) image difference captioning restricted to camera movement and viewpoint changes. Macro-averaged F1* in percent, combining the share of true differences described and the share of described differences that are correct, as judged by Gemini 2.5 Pro; higher is better.

Metric: Macro-averaged F1* (%) on camera movement and viewpoint 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: Around February 2027. 14 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1GPT-5.289.4#185
2Gemini 3 Pro86.5#145
3Step3 VL 10B77.5#1080
4Qwen 3 VL Plus75.1#512
5Kimi K2.5Thinking74.5#266
6Grok 4.2072.7#301
7MiMo-VL-7B-RL-250872.7#823
8Qwen 3 VL 32B Instruct71.7#606
9Qwen 3 VL 8B Instruct70.3#1029
10Seed1.6-VLThinking65.3#385
11InternVL3.5-8B64.9#1604
12Qwen 3 VL 32B (Thinking)63.3#573
13Qwen 3 VL 8B (Thinking)62.9#910
14InternVL3.5-38B57.2#1021

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=diffcap-bench-camera-and-viewpoint · How It Works · Data refreshed daily, snapshot 2026-11-21.