DeckBench: Deck rating

Top 10 of 30 models · rating, higher is better · lines show the 95% range · ≈ cannot be told apart from the leader

  1. 1≈ GPT-6 AstraOpenAI 1,588 (95% range 1,491–1,775)
  2. 2≈ GPT-6 SolOpenAI 1,470 (95% range 1,405–1,584)
  3. 3≈ Claude Fable 5.2Anthropic 1,419 (95% range 1,315–1,559)
  4. 4≈ GPT-6.1 SolOpenAI 1,416 (95% range 1,351–1,517)
  5. 5 Claude Opus 5.5Anthropic 1,371 (95% range 1,293–1,468)
  6. 6 Claude Sonnet 5.5Anthropic 1,348 (95% range 1,268–1,443)
  7. 7 GPT-6.1 TerraOpenAI 1,327 (95% range 1,219–1,436)
  8. 8 GPT-6 LunaOpenAI 1,287 (95% range 1,220–1,372)
  9. 9 Claude Fable 5.1Anthropic 1,278 (95% range 1,222–1,365)
  10. 10 MiniMax M3.1 FlashMiniMax 1,277 (95% range 1,196–1,405)

Results to 10 Nov 2026 · release 8342a7178929Source: Spring Prompt · CC BY 4.0