{
  "generated": "2026-09-08",
  "benchmark": "cybench",
  "results": [
    {
      "system": "Claude 3.5 Sonnet",
      "developer": "Anthropic",
      "value": 17.5,
      "date": "2024-08-15",
      "source": {
        "url": "https://arxiv.org/abs/2408.08926",
        "kind": "paper",
        "accessed": "2026-09-04"
      },
      "conditions": {
        "scaffold": "Cybench agent (structured bash)",
        "notes": "Paper release; 7/40 tasks unguided; subtask-guided 15%, 43.9% of subtasks solved."
      }
    },
    {
      "system": "GPT-4o",
      "developer": "OpenAI",
      "value": 12.5,
      "date": "2024-08-15",
      "source": {
        "url": "https://arxiv.org/abs/2408.08926",
        "kind": "paper",
        "accessed": "2026-09-04"
      },
      "conditions": {
        "scaffold": "Cybench agent (structured bash)",
        "notes": "Paper release; 5/40 tasks unguided; subtask-guided 17.5%."
      }
    },
    {
      "system": "Claude Opus 4.5",
      "developer": "Anthropic",
      "value": 82.0,
      "date": "2025-11-24",
      "source": {
        "url": "https://cybench.github.io",
        "kind": "developer-report",
        "accessed": "2026-09-04"
      },
      "conditions": {
        "notes": "Claude Opus 4.5 System Card number republished on the official leaderboard (footnote 3): 39-task subset, average pass@1. Dated to the model's release day."
      }
    },
    {
      "system": "Claude Mythos Preview",
      "developer": "Anthropic",
      "value": 100.0,
      "date": "2026-04-07",
      "source": {
        "url": "https://cybench.github.io",
        "kind": "developer-report",
        "accessed": "2026-09-04"
      },
      "conditions": {
        "notes": "Claude Mythos Preview System Card number republished on the official leaderboard (footnote 6): 35-task subset. Dated to the model's announcement day."
      }
    },
    {
      "system": "Claude Opus 4.7",
      "developer": "Anthropic",
      "value": 96.0,
      "date": "2026-04-16",
      "source": {
        "url": "https://cybench.github.io",
        "kind": "developer-report",
        "accessed": "2026-09-04"
      },
      "conditions": {
        "notes": "Claude Opus 4.7 System Card number republished on the official leaderboard (footnote 8): 35-task subset. Dated to the model's release day."
      }
    }
  ]
}