HELM VHELM - Seed Bench - Instance Interaction: leaderboard

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

Top models

#ModelScore
1O3 (2025-04-16)86.6
2O1 (2024-12-17)86.6
3GPT-4.586.6
4GPT-4.1 Mini85.57
5Claude 3.5 Sonnet (20241022)84.54
6Gemini 2.5 Pro (Preview 03-25)84.54
7GPT-4 Turbo84.54
8Gemini 1.5 Flash (002)83.51
9GPT-4o (2024-05-13)82.47
10GPT-4o (2024-11-20)82.47
11O4 Mini (2025-04-16)81.44
12GPT-4.1 (2025-04-14)80.41
13GPT-4o (2024-08-06)80.41
14Gemini 2.0 Flash (Preview)80.41
15Gemini 1.5 Pro (002)79.38

Interactive version: theaggregate.ai/benchmark?slug=helm-vhelm-seed-bench-instance-interaction · How It Works · Data refreshed daily, snapshot 2026-09-19.