Overall

Models play 30 unseen puzzles and are scored 0–100 across Xiangqi, Chess, Go, Backgammon, Shogi, Scrabble, Kalah, Gomoku, Othello, and Connect Four.

Quality vs cost

8 of 8 models
  • Anthropic
  • OpenAI
  • DeepSeek
  • Z.ai
  • Google
404550556065707580$0.001$0.01$0.1$1Claude Haiku 5.5 High · 39.0% · <$0.01 · Pareto frontierClaude Opus 5.5 High · 37.6% · $0.08DeepSeek V4.1 Flash High · 39.9% · $0.03Gemini 4 Argon High · 66.3% · $0.28 · Pareto frontierGLM-5.3 Flash High · 43.9% · $0.01 · Pareto frontierGPT-6 Astra High · 67.2% · $0.56 · Pareto frontierGPT-6 Luna High · 42.5% · <$0.01 · Pareto frontierGPT-6.1 Sol High · 56.4% · $0.10 · Pareto frontierGPT-6 Astra HighGemini 4 Argon HighGPT-6.1 Sol HighGLM-5.3 Flash HighGPT-6 Luna HighDeepSeek V4.1 Flash HighClaude Haiku 5.5 HighClaude Opus 5.5 High
  • Pareto frontier: nothing plotted is both cheaper and better
  • Anthropic
  • OpenAI
  • DeepSeek
  • Z.ai
  • Google

Horizontal axis is logarithmic.

Look