MEGA-Bench Task - Semantic Matching Of Two Images: leaderboard

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

Top models

#ModelScore
1GPT-4o64.3
2Claude 3.5 Sonnet (20241022)50
3Claude 3.5 Sonnet (20240620)50
4Gemma 3 12B (IT)42.9
5Gemini 2.5 Pro (Preview 03-25)42.9
6Gemma 3 27B (IT)35.7
7InternVL3-78B35.7
8Gemini 1.5 Pro (002)35.7
9InternVL3-38B28.6
10Pixtral-12B28.6
11InternVL2.5-2B28.6
12InternVL3-8B21.4
13Qwen 2 VL 72B21.4
14InternVL2-8B14.3
15Gemini 1.5 Flash (002)14.3

Interactive version: theaggregate.ai/benchmark?slug=mega-bench-task-semantic-matching-of-two-images · How It Works · Data refreshed daily, snapshot 2026-09-05.