{
  "api_version": "1",
  "updated": "2026-10-25",
  "latest_event": "2026-10-24",
  "rights": "© Frontier Log, all rights reserved.",
  "benchmark": {
    "slug": "cybergym",
    "name": "CyberGym",
    "family": "cybergym",
    "unit": "%",
    "higher_is_better": true,
    "status": "active",
    "active": true,
    "superseded_by": null,
    "benchmark_url": "https://www.cybergym.io",
    "introduced": "2025-06-03",
    "creator": "UC Berkeley",
    "introduced_source": "https://arxiv.org/abs/2506.02548",
    "settings": [
      {
        "setting": "default",
        "name": "",
        "url": "https://frontierlog.ai/benchmark/cybergym",
        "results": 4
      }
    ]
  },
  "total": 4,
  "results": [
    {
      "id": "glm-5-3:cybergym:default:1",
      "model": "glm-5-3",
      "model_name": "GLM-5.3",
      "model_date": "2026-08-14",
      "event": "https://frontierlog.ai/event/2026-08-14-glm-5-3",
      "benchmark": "cybergym",
      "benchmark_name": "CyberGym",
      "setting": "default",
      "setting_name": "",
      "score": 85.0,
      "unit": "%",
      "printed": "85%",
      "reported_by": "zai",
      "run_by": "Anthropic",
      "label": "GLM-5.3, model",
      "source": "https://www.anthropic.com/research/glm-5-3-and-the-spread-of-advanced-cyber-capabilities",
      "source_title": "GLM-5.3 and the spread of advanced cyber capabilities",
      "source_date": "2026-09-29",
      "where": "Chart 'Abliteration removes GLM-5.3's refusals and leaves its capabilities largely intact', lower panel 'Capability benchmarks, before and after abliteration', CyberGym (subset) group, GLM-5.3 'model' bar value label",
      "own": false,
      "primary": false,
      "parallel": false,
      "notes": "185-task subset; paired with Anthropic's abliterated copy",
      "caveat": "",
      "human_verified": true,
      "human_verified_on": "2026-10-25",
      "check": "human",
      "corroborated": false,
      "corroborated_on": null,
      "corroborated_by": [],
      "source_image": "https://frontierlog.ai/sources/5ecd061698037a5e.webp",
      "source_image_original": "https://cdn.sanity.io/images/4zrzovbb/website/d4d48dcf67d529d3b2243bfa50a7cccc98565278-1280x1531.webp"
    },
    {
      "id": "glm-5-3-abliterated-anthropic:cybergym:default:1",
      "model": "glm-5-3-abliterated-anthropic",
      "model_name": "GLM-5.3 abliterated (Anthropic)",
      "model_date": "2026-08-14",
      "event": "https://frontierlog.ai/event/2026-08-14-glm-5-3",
      "benchmark": "cybergym",
      "benchmark_name": "CyberGym",
      "setting": "default",
      "setting_name": "",
      "score": 81.0,
      "unit": "%",
      "printed": "81%",
      "reported_by": "zai",
      "run_by": "Anthropic",
      "label": "GLM-5.3, abliterated",
      "source": "https://www.anthropic.com/research/glm-5-3-and-the-spread-of-advanced-cyber-capabilities",
      "source_title": "GLM-5.3 and the spread of advanced cyber capabilities",
      "source_date": "2026-09-29",
      "where": "Chart 'Abliteration removes GLM-5.3's refusals and leaves its capabilities largely intact', lower panel 'Capability benchmarks, before and after abliteration', CyberGym (subset) group, GLM-5.3 'abliterated' bar value label",
      "own": false,
      "primary": true,
      "parallel": false,
      "notes": "185-task subset",
      "caveat": "",
      "human_verified": true,
      "human_verified_on": "2026-10-25",
      "check": "human",
      "corroborated": false,
      "corroborated_on": null,
      "corroborated_by": [],
      "source_image": "https://frontierlog.ai/sources/5ecd061698037a5e.webp",
      "source_image_original": "https://cdn.sanity.io/images/4zrzovbb/website/d4d48dcf67d529d3b2243bfa50a7cccc98565278-1280x1531.webp"
    },
    {
      "id": "glm-5-3-flash:cybergym:default:1",
      "model": "glm-5-3-flash",
      "model_name": "GLM-5.3-Flash",
      "model_date": "2026-08-26",
      "event": "https://frontierlog.ai/event/2026-08-26-glm-5-3-flash",
      "benchmark": "cybergym",
      "benchmark_name": "CyberGym",
      "setting": "default",
      "setting_name": "",
      "score": 86.0,
      "unit": "%",
      "printed": "86%",
      "reported_by": "zai",
      "run_by": "Anthropic",
      "label": "GLM-5.3-Flash, model",
      "source": "https://www.anthropic.com/research/glm-5-3-and-the-spread-of-advanced-cyber-capabilities",
      "source_title": "GLM-5.3 and the spread of advanced cyber capabilities",
      "source_date": "2026-09-29",
      "where": "Chart 'Abliteration removes GLM-5.3's refusals and leaves its capabilities largely intact', lower panel 'Capability benchmarks, before and after abliteration', CyberGym (subset) group, GLM-5.3-Flash 'model' bar value label",
      "own": false,
      "primary": false,
      "parallel": false,
      "notes": "95-task subset; paired with Anthropic's abliterated copy",
      "caveat": "",
      "human_verified": true,
      "human_verified_on": "2026-10-25",
      "check": "human",
      "corroborated": false,
      "corroborated_on": null,
      "corroborated_by": [],
      "source_image": "https://frontierlog.ai/sources/5ecd061698037a5e.webp",
      "source_image_original": "https://cdn.sanity.io/images/4zrzovbb/website/d4d48dcf67d529d3b2243bfa50a7cccc98565278-1280x1531.webp"
    },
    {
      "id": "glm-5-3-flash-abliterated-anthropic:cybergym:default:1",
      "model": "glm-5-3-flash-abliterated-anthropic",
      "model_name": "GLM-5.3-Flash abliterated (Anthropic)",
      "model_date": "2026-08-26",
      "event": "https://frontierlog.ai/event/2026-08-26-glm-5-3-flash",
      "benchmark": "cybergym",
      "benchmark_name": "CyberGym",
      "setting": "default",
      "setting_name": "",
      "score": 86.0,
      "unit": "%",
      "printed": "86%",
      "reported_by": "zai",
      "run_by": "Anthropic",
      "label": "GLM-5.3-Flash, abliterated",
      "source": "https://www.anthropic.com/research/glm-5-3-and-the-spread-of-advanced-cyber-capabilities",
      "source_title": "GLM-5.3 and the spread of advanced cyber capabilities",
      "source_date": "2026-09-29",
      "where": "Chart 'Abliteration removes GLM-5.3's refusals and leaves its capabilities largely intact', lower panel 'Capability benchmarks, before and after abliteration', CyberGym (subset) group, GLM-5.3-Flash 'abliterated' bar value label",
      "own": false,
      "primary": true,
      "parallel": false,
      "notes": "95-task subset",
      "caveat": "",
      "human_verified": true,
      "human_verified_on": "2026-10-25",
      "check": "human",
      "corroborated": false,
      "corroborated_on": null,
      "corroborated_by": [],
      "source_image": "https://frontierlog.ai/sources/5ecd061698037a5e.webp",
      "source_image_original": "https://cdn.sanity.io/images/4zrzovbb/website/d4d48dcf67d529d3b2243bfa50a7cccc98565278-1280x1531.webp"
    }
  ]
}