{
  "arena-agent.json": {
    "src": "https://arena.ai/leaderboard/agent 2026-08-17 抓取，1,793,983 sessions / 49 models；值为该模型最好一档的**排名**（1=最好）"
  },
  "arena-elo-for-swebp.json": {
    "src": "arena.ai/leaderboard/text 2026-08-17；映射表写死于抓取脚本，22/22 全配上，无手工牵线"
  },
  "arena-text.json": {
    "src": "https://arena.ai/leaderboard/text 2026-08-17 抓取，7,779,985 votes / 391 models；值为该模型在 Arena 上最好一档的 Elo"
  },
  "deepswe-v1.1.json": {
    "src": "https://deepswe.datacurve.ai/  (Leaderboard → v1.1 → All effort levels → Models: Select all)"
  },
  "rli.json": {
    "src": "https://safe.ai/blog/significant-increase-in-digital-labor-automation 2026-08-17；automation rate %"
  },
  "sweatlas-qna.json": {
    "src": "https://labs.scale.com/leaderboard/sweatlas-qna 2026-08-17 抓取。124 题，Task Resolve Rate（严格阈值 1.0，LLM 判官 Claude Opus 4.5 逐条打鲁棒分）"
  },
  "swebench-pro-public.json": {
    "src": "https://labs.scale.com/leaderboard/swe_bench_pro_public 2026-08-17；25 条"
  },
  "terminal-bench-2.1.json": {
    "src": "https://www.tbench.ai/leaderboard/terminal-bench/2.1 2026-08-17；17 条；带 harness、effort、成本"
  }
}