All benchmarks

GDPval-AA v2.1

Knowledge work

GDPval-AA is Artificial Analysis’s independent run of GDPval, which measures model performance on real, economically valuable knowledge-work deliverables across dozens of occupations. Version 2.1 uses 220 tasks across 44 occupations and nine industries, with models given shell access and web browsing in an agentic loop. Results are an Elo rating derived from blind pairwise comparisons, anchored to DeepSeek V4.1 Flash at 1600 — not a percentage.

Model scores

  • Fable 5.21831
  • Mythos 5.2—
  • Fable 5.11763
  • Fable 51618
  • Opus 5.51866
  • Opus 51730
  • Opus 4.81462
  • Sonnet 5.51838
  • Sonnet 51470
  • GPT-6 Astra1586
  • GPT-6.1 Sol1608
  • GPT-5.6 Sol1618
  • Grok 4.71717
  • GPT-5.51359
  • Composer 2.5—
  • Opus 4.71362
  • Gemini 3.1 Pro799
  • Mythos Preview—

Official source: Artificial Analysis — GDPval-AA v2.1 leaderboard

Related reading