HELM Safety Anthropic Red Team: leaderboard

Metric: LM Evaluated Safety score (%). Source: storage.googleapis.com. 87 models tracked.

Top models

#ModelScore
1Claude 3 Haiku 20240307100
2Gemini 1.5 Flash 00199.9
3Gemini 1.5 Pro 00199.9
4IBM Granite 4.0 Small with guardian99.9
5Qwen3 235B A22B Instruct 2507 FP899.9
6Qwen3-Next 80B A3B Thinking99.9
7Claude 3.5 Sonnet 2024062099.8
8Claude 3 Opus 2024022999.8
9Claude 3 Sonnet 2024022999.8
10Claude 3.7 Sonnet 2025021999.7

Interactive version: theaggregate.ai/benchmark?slug=helm-safety-anthropic-red-team-1ttfm3b · How It Works · Data refreshed daily, snapshot 2026-10-09.