Frontier Security Agents - Cybench Hard — leaderboard
Metric: Success rate (%). Source: arxiv.org. 8 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.5 | 94.1 |
| 2 | DeepSeek V4 Flash | 86.4 |
| 3 | GPT-5.6 Luna | 79.5 |
| 4 | Claude Opus 4.8 | 76.2 |
| 5 | GPT-5.6 Terra | 65.8 |
| 6 | DeepSeek V4 Pro | 43.9 |
| 7 | GPT-5.6 Sol | 9.4 |
| 8 | Claude Fable 5 | 0 |
Interactive version: aibenchmarks.dev/benchmark?slug=frontier-security-agents-cybench-hard · How the rankings work · Data refreshed daily, snapshot 2026-07-18.