HELM Arabic - Aratrust - Illegal: leaderboard

Metric: Exact Match (%). Source: crfm.stanford.edu. 41 models tracked.

Top models

#ModelScore
1Claude Sonnet 4.698.11
2Gemma 4 31B (IT)98.11
3Claude Opus 4.798.11
4Mistral Large 398.11
5Gemini 2.5 Flash (Thinking)98.11
6Gemini 2.5 Flash Lite (Thinking)98.11
7GPT-4.1 (2025-04-14)96.23
8Claude Haiku 4.5 (20251001)96.23
9Llama 4 Maverick Instruct FP896.23
10Mistral Large 2 (Nov) Instruct (2411)96.23
11AceGPT-v2-70B-Chat96.23
12GPT-5.4 (2026-03-05)96.23
13Qwen 2.5 72B Instruct94.34
14Llama 4 Scout Instruct94.34
15DeepSeek V3.194.34

Interactive version: theaggregate.ai/benchmark?slug=helm-arabic-aratrust-illegal · How It Works · Data refreshed daily, snapshot 2026-09-19.