data/leaderboard.json

{
  "updated": "2026-06-10T18:19:04.013Z",
  "sources": {
    "eog": {
      "url": "https://raw.githubusercontent.com/EnterpriseOps-Gym/EnterpriseOps-Gym.github.io/main/script.js",
      "hash": "sha256:f96e11145e82ffd4bb2654a9e7cf42f5ebd3b10f481aa1027be22a0d3a1badba",
      "fetchedAt": "2026-06-10T18:19:04.012Z"
    },
    "eva": {
      "url": "https://raw.githubusercontent.com/ServiceNow/eva/main/website/src/data/leaderboardStats.json",
      "hash": "sha256:1f7a33021ffea8a3719c24cf31f75cfeff4b0271f77ef8abfd0063099081b288",
      "fetchedAt": "2026-06-10T18:19:04.013Z"
    }
  },
  "eog": {
    "metricLabel": "Task Success Rate · Oracle mode",
    "rows": [
      {
        "rank": 1,
        "model": "Gemini 3.5 Flash (High)",
        "org": "Google",
        "type": "closed",
        "score": 46,
        "bar": 46
      },
      {
        "rank": 2,
        "model": "Claude Opus 4.6",
        "org": "Anthropic",
        "type": "closed",
        "score": 44.6,
        "bar": 45
      },
      {
        "rank": 3,
        "model": "GPT-5.5 (High)",
        "org": "OpenAI",
        "type": "closed",
        "score": 41.2,
        "bar": 41
      },
      {
        "rank": 4,
        "model": "Claude Opus 4.7 (High)",
        "org": "Anthropic",
        "type": "closed",
        "score": 41.2,
        "bar": 41
      },
      {
        "rank": 5,
        "model": "Claude Sonnet 4.6",
        "org": "Anthropic",
        "type": "closed",
        "score": 40.4,
        "bar": 40
      }
    ]
  },
  "evaAccuracy": {
    "metricLabel": "Pass@1",
    "rows": [
      {
        "rank": 1,
        "name": "Scribe v2.2 Realtime + GPT-5.4 + Eleven Flash v2 (ElevenAgents)",
        "subtitle": "Mixed Models · Cascade",
        "score": 0.67,
        "ciLower": 0.62,
        "ciUpper": 0.72,
        "bar": 67
      },
      {
        "rank": 2,
        "name": "Grok Voice Think Fast 1.0",
        "subtitle": "Grok · Speech-to-Speech",
        "score": 0.59,
        "ciLower": 0.54,
        "ciUpper": 0.64,
        "bar": 59
      },
      {
        "rank": 3,
        "name": "Nova 3 + GPT-5.4 + Sonic 3",
        "subtitle": "Mixed Models · Cascade",
        "score": 0.5,
        "ciLower": 0.46,
        "ciUpper": 0.54,
        "bar": 50
      }
    ]
  },
  "evaExperience": {
    "metricLabel": "Pass@1",
    "rows": [
      {
        "rank": 1,
        "name": "Scribe v2.2 Realtime + Claude Haiku 4.5 + Eleven Flash v2 (ElevenAgents)",
        "subtitle": "Mixed Models · Cascade",
        "score": 0.82,
        "ciLower": 0.79,
        "ciUpper": 0.84,
        "bar": 82
      },
      {
        "rank": 2,
        "name": "Gemini 3.1 Flash Live",
        "subtitle": "Google · Speech-to-Speech",
        "score": 0.59,
        "ciLower": 0.56,
        "ciUpper": 0.62,
        "bar": 59
      },
      {
        "rank": 3,
        "name": "Grok Voice Think Fast 1.0",
        "subtitle": "Grok · Speech-to-Speech",
        "score": 0.57,
        "ciLower": 0.53,
        "ciUpper": 0.61,
        "bar": 57
      }
    ]
  }
}