CyberSecEval-3 — leaderboard

Meta's cybersecurity risk assessment evaluating LLMs across 7 risk dimensions: social engineering, vulnerability exploitation, prompt injection, insecure code generation, and more.

Metric: Score (%). Source: huggingface.co. 6 models tracked.

Top models

#ModelScore
1Llama 3 70B44.14
2Mixtral 8x22B37.51
3GPT-4 Turbo34.94
4Gemini 1.0 Pro34.5
5Qwen 2 72B Instruct34.4
6Llama 3.1 405B33.14

Interactive version: theaggregate.ai/benchmark?slug=cyberseceval-3 · How the rankings work · Data refreshed daily, snapshot 2026-07-22.