ArtificialAnalysis/Coding Agent Index

• Artificial Analysis Coding Agent Benchmarks

  • Composite score: average pass@1 across three benchmarks (three runs each).
  • Components:
    • SWE-Bench-Pro-Hard-AA — code generation, 150 tasks
    • Terminal-Bench v2 — agentic terminal use, 84 tasks
    • SWE-Atlas-QnA — technical Q&A, 124 tasks
  • Headline metrics on the leaderboard:
    • ArtificialAnalysis/Coding Agent Index/Cost per Task