AI-modeller
Alla topplistor
Agenter & datoranvändning · Artificial Analysis

GDPval-AA v2.1

Hur bra AI-agenter är på riktiga arbetsuppgifter från 44 yrken, som rapporter, kalkylark och presentationer, bedömt genom blinda jämförelser mellan modellernas leveranser.

Placeringar (27, varav 25 i vår lista)

  1. 1
    Anthropic Claude Opus 5.5 (max with fallback)
    max
    1866 (1840–1892)[1]
  2. 2
    Meta Muse Wonder Full (max)
    Muse Wonder Full, max
    1857 (1832–1882)[1]
  3. 3
    Moonshot AI Kimi K3.1 Pro
    1846 (1820–1872)[1]
  4. 4
    Anthropic Claude Sonnet 5.5 (max with fallback)
    max
    1838 (1814–1861)[1]
  5. 5
    Meta Muse Wonder (max)
    Muse Wonder (inte Full), max
    1770 (1747–1793)[1]
  6. 6
    Anthropic Claude Fable 5.1 (max with fallback)
    max
    1755 (1735–1774)[1]
  7. 7
    DeepSeek DeepSeek V4.1 Flash 1017 (max)
    max
    1746 (1725–1767)[1]
  8. 8
    xAI Grok 4.7 (xhigh)
    xhigh
    1714 (1692–1736)[1]
  9. 9
    Xiaomi MiMo-V2.6-Pro
    Pro
    1683 (1662–1704)[1]
  10. 10
    Alibaba (Qwen, Wan, Tongyi) Qwen3.8 Max (0902)
    1666 (1646–1686)[1]
  11. 11
    Z.ai (Zhipu) GLM-5.3 (max)
    max
    1649 (1630–1668)[1]
  12. 12
    Z.ai (Zhipu) GLM-5.3-Flash
    1642 (1625–1660)[1]
  13. 13
    Google Gemini 4 Argon (high)
    high
    1624 (1603–1645)[1]
  14. 14
    Anthropic Claude Haiku 5.5 (max)
    max
    1619 (1597–1642)[1]
  15. 15
    OpenAI GPT-6.1 Sol (max)
    max
    1596 (1574–1619)[1]
  16. 16
    StepFun Step 5 Preview
    1587 (1563–1611)[1]
  17. 17
    OpenAI GPT-6 Astra (max)
    max
    1574 (1555–1594)[1]
  18. 18
    MiniMax MiniMax-M3.1-Flash
    1458 (1434–1481)[1]
  19. 19
    Google Gemini 3.8 Flash (high)
    high
    1436 (1417–1456)[1]
  20. 20
    OpenAI GPT-6 Luna (max)
    max
    1433 (1412–1454)[1]
  21. 21
    Mistral AI Mistral Large 4 Preview
    förhandsversion
    1422 (1398–1446)[1]
  22. 22
    Alibaba (Qwen, Wan, Tongyi) Qwen3.8 27B (xhigh)
    xhigh
    1419 (1399–1439)[1]
  23. 23
    Meta Muse Spark Open (xhigh)
    xhigh
    1401 (1382–1420)[1]
  24. 24
    MBZUAI K2 Horizon 375B A23B
    1358 (1339–1378)[1]
  25. 25
    Beam 1.0
    Reflection
    1177 (1158–1196)[1]
  26. 26
    NVIDIA Nemotron 3.5 Super VL
    1162 (1144–1181)[1]
  27. 27
    Inkling (xhigh)
    Thinking Machines · xhigh
    1072 (1050–1094)[1]

Källor

  1. [1]
    GDPval-AA v2.1 Benchmark ↗
    Artificial AnalysisTopplistaartificialanalysis.ai
    Publicerad okänt datum Läst 25 okt 2026
  2. [2]
    GDPval: Evaluating AI Model Performance on Real-World Economically Valuable Tasks ↗
    OpenAIForskningsartikelarxiv.org
    Publicerad okänt datum Läst 25 okt 2026

Historik

  1. 25 okt 2026
    Tillagd. Avläst 25 okt 2026 (GDPval-AA v2.1).