Cyber Bench Hub (v2): leaderboard
Priam AI's evaluation of models on public cyber-security datasets. Only the v2 column is tracked; v1 sits at 99-100% for almost every model and carries no ranking signal. Even v2 runs near the ceiling, so treat small gaps here as noise.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 22 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.0 Flash | 99 | #342 |
| 2 | DeepSeek R1 | 98 | #251 |
| 3 | Claude 3.7 Sonnet | 98 | #249 |
| 4 | Phi-3 Mini 128K Instruct | 98 | #1298 |
| 5 | Llama 3.3 70B Instruct | 97 | #534 |
| 6 | GPT-4o Mini | 97 | #599 |
| 7 | DeepSeek V3 | 97 | #321 |
| 8 | Claude 3.5 Sonnet | 97 | #350 |
| 9 | Phi-4 | 97 | #706 |
| 10 | Mistral Small 3.1 | 97 | #610 |
| 11 | Mistral Small 3 | 97 | #627 |
| 12 | Claude 3.5 Haiku | 97 | #568 |
| 13 | Gemini 2.0 Flash Lite | 97 | #450 |
| 14 | Phi-3.5-mini-instruct | 97 | #1167 |
| 15 | GPT-4.5 (Preview) | 97 | #244 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=cyber-bench-hub-v2 · How It Works · Data refreshed daily, snapshot 2026-10-20.