CyberSecEval-3 Insecure Autocomplete: leaderboard
Metric: Insecure code generation (%). Source: arxiv.org. 6 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 1.0 Pro | 25 |
| 2 | Mixtral 8x22B | 25 |
| 3 | Llama 3 70B | 28 |
| 4 | GPT-4 Turbo | 30 |
| 5 | Qwen 2 72B Instruct | 30 |
| 6 | Llama 3.1 405B | 31 |
Interactive version: theaggregate.ai/benchmark?slug=cyberseceval-3-insecure-autocomplete · How It Works · Data refreshed daily, snapshot 2026-10-09.