HELM VHELM - Blink Perception - Semantic Correspondence: leaderboard

Metric: Prefix Quasi-Exact Match (%). Source: crfm.stanford.edu. 48 models tracked.

Top models

#ModelScore
1O3 (2025-04-16)83.45
2O1 (2024-12-17)78.42
3O4 Mini (2025-04-16)77.7
4GPT-4.576.98
5GPT-4.1 (2025-04-14)74.1
6Gemini 2.0 Flash Lite74.1
7Gemini 2.5 Pro (Preview 03-25)73.38
8Gemini 2.0 Flash (Preview)73.38
9Gemini 1.5 Pro (002)71.22
10Gemini 2.0 Pro (Preview 02-05)71.22
11Gemini 2.0 Flash68.35
12GPT-4o (2024-08-06)68.35
13GPT-4o (2024-05-13)66.91
14Llama 4 Maverick Instruct FP866.91
15Claude 3.5 Sonnet (20240620)65.47

Interactive version: theaggregate.ai/benchmark?slug=helm-vhelm-blink-perception-semantic-correspondence · How It Works · Data refreshed daily, snapshot 2026-09-19.