data/leaderboard.json

{
  "updated": "2026-07-17T21:00:09.508Z",
  "sources": {
    "eog": {
      "url": "https://artificialanalysis.ai/evaluations/enterprise-ops-gym-aa/",
      "hash": "sha256:617ad42070acaf5d70ca7782c7f1e4960d26f56cd228667a86f89d51abc580c8",
      "fetchedAt": "2026-07-17T21:00:09.508Z"
    },
    "eva": {
      "url": "https://raw.githubusercontent.com/ServiceNow/eva/main/website/src/data/leaderboardStats.json",
      "hash": "sha256:e71bd9e18c294322e706ebc9ca8ac477be01c8caf5d9c25b3121566753f35b77",
      "fetchedAt": "2026-07-17T21:00:09.508Z"
    }
  },
  "eog": {
    "metricLabel": "Task Success Rate · Oracle mode",
    "rows": [
      {
        "rank": 1,
        "model": "Claude Fable 5 (with fallback)",
        "org": "Anthropic",
        "score": 51.1,
        "bar": 51
      },
      {
        "rank": 2,
        "model": "Gemini 3.5 Flash",
        "org": "Google",
        "score": 50.1,
        "bar": 50
      },
      {
        "rank": 3,
        "model": "GPT-5.5 (xhigh)",
        "org": "OpenAI",
        "score": 46.6,
        "bar": 47
      },
      {
        "rank": 4,
        "model": "Kimi K3",
        "org": "Kimi",
        "score": 45.3,
        "bar": 45
      },
      {
        "rank": 5,
        "model": "Qwen3.7 Max",
        "org": "Alibaba",
        "score": 45,
        "bar": 45
      }
    ]
  },
  "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": "Universal 3.5 Pro + GPT-5.4 + Sonic 3.5",
        "subtitle": "Mixed Models · Cascade",
        "score": 0.6,
        "ciLower": 0.55,
        "ciUpper": 0.65,
        "bar": 60
      },
      {
        "rank": 3,
        "name": "Grok Voice Think Fast 1.0",
        "subtitle": "Grok · Speech-to-Speech",
        "score": 0.59,
        "ciLower": 0.54,
        "ciUpper": 0.64,
        "bar": 59
      }
    ]
  },
  "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
      }
    ]
  }
}