MEGA-Bench Task - Visual Prediction Rater Surface Normal Estimation: leaderboard

Metric: Task Score (%). Source: huggingface.co. 44 models tracked.

Top models

#ModelScore
1Gemini 2.5 Pro (Preview 03-25)90.5
2Gemini 1.5 Pro (002)81
3InternVL3-38B73.8
4InternVL3-78B66.7
5Claude 3.5 Sonnet (20241022)66.7
6GPT-4o57.1
7Claude 3.5 Sonnet (20240620)57.1
8InternVL2.5-78B54.8
9Gemma 3 27B (IT)42.9
10Gemini 1.5 Flash (002)33.3
11Aquila-VL-2B26.2
12Qwen 2 VL 72B16.7
13Pixtral-12B16.7
14Llama 4 Scout Base15.5
15Gemma 3 4B (IT)11.9

Interactive version: theaggregate.ai/benchmark?slug=mega-bench-task-visual-prediction-rater-surface-normal-estimation · How It Works · Data refreshed daily, snapshot 2026-09-05.