DistortBench - Luminance: leaderboard

DistortBench (2026) shows a vision-language model a distorted photo without its original and asks it to name the distortion type and severity among four options (chance 25 percent). Joint accuracy on the four luminance distortions (brighten, darken, mean shift, contrast), in percent; higher is better.

Metric: Joint accuracy (%) on the four luminance distortions (brighten, darken, mean shift, contrast), a category of DistortBench's 13,500 four-choice questions (27 distortion types at five KADID-10k calibrated severity levels on 100 KADIS-700k photos, no reference image), the answer naming both type and severity, answered-only accuracy (unparseable responses, at most 3.4 percent, leave the denominator), chance 25; higher is better. Source: arxiv.org. Saturation forecast: Around April 2027. 14 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1Qwen 3.5 27BNon-reasoning59.2#375
2Qwen 3.5 35B A3BNon-reasoning51.5#492
3GPT-5.451.4#135
4Qwen 3.5 4BNon-reasoning45.7#1241
5Kimi-VL-A3B-Instruct45.7#2253
6Qwen 3.5 9BNon-reasoning43.8#828
7InternVL3.5-38B43.6#1021
8Gemma 3 27B40.5#1663
9Kimi-VL-A3B (Thinking)38.8
10Gemma 3 12B38#2059
11InternVL3.5-14B37.2#1504
12InternVL3.5-8B36.4#1604
13Qwen 2 VL 7B34.7#2275
14Gemma 3 4B19.4#5339

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

Interactive version: theaggregate.ai/benchmark?slug=distortbench-luminance · How It Works · Data refreshed daily, snapshot 2026-11-21.