Connect Four

7×6 Connect Four, given as the columns played. An exact solver scores every column by win, draw or loss and how soon; the answer is a column. See Methodology.

Quality is the mean oracle score of the moves played, × 100. The best move scores 100; an illegal move or no move scores 0. Protocol boardbench-0.2.

Quality vs cost

7 of 7 models
  • Anthropic
  • OpenAI
  • DeepSeek
  • Z.ai
  • Meta
2030405060708090100$0.001$0.01$0.1$1Claude Haiku 5.5 High · 80.4% · <$0.01 · Pareto frontierClaude Opus 5.5 High · 47.1% · $0.08DeepSeek V4.1 Flash High · 33.3% · $0.04GLM-5.3 Flash High · 66.7% · $0.03GPT-6 Luna High · 66.7% · $0.01GPT-6.1 Sol High · 66.7% · $0.10Muse Spark 1.3 High · 13.8% · $0.15Claude Haiku 5.5 HighGPT-6 Luna HighGLM-5.3 Flash HighGPT-6.1 Sol HighClaude Opus 5.5 HighDeepSeek V4.1 Flash HighMuse Spark 1.3 High
  • Pareto frontier: nothing plotted is both cheaper and better
  • Anthropic
  • OpenAI
  • DeepSeek
  • Z.ai
  • Meta

Horizontal axis is logarithmic.

Table

ModelQualityLegalityCostn
Haiku 5.5High80.4100.0<$0.013
GLM-5.3 FlashHigh66.7100.0$0.033
LunaHigh66.7100.0$0.013
SolHigh66.7100.0$0.103
Opus 5.5High47.1100.0$0.083
DeepSeek V4.1 FlashHigh33.3100.0$0.043
Muse Spark 1.3High13.866.7$0.153

Headline is the mean move score × 100. n is answers: puzzles × samples.

Look