Shogi

Standard shogi, given as SFEN. YaneuraOu scores every legal move, drops included; the answer is a USI move. 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 · 22.4% · <$0.01 · Pareto frontierClaude Opus 5.5 High · 45.6% · $0.04DeepSeek V4.1 Flash High · 40.4% · $0.03GLM-5.3 Flash High · 45.6% · $0.02GPT-6 Luna High · 55.8% · <$0.01 · Pareto frontierGPT-6.1 Sol High · 100.0% · $0.08 · Pareto frontierMuse Spark 1.3 High · 53.5% · $0.13GPT-6.1 Sol HighGPT-6 Luna HighMuse Spark 1.3 HighGLM-5.3 Flash HighClaude Opus 5.5 HighDeepSeek V4.1 Flash HighClaude Haiku 5.5 High
  • Pareto frontier: nothing plotted is both cheaper and better
  • Anthropic
  • OpenAI
  • DeepSeek
  • Z.ai
  • Meta

Horizontal axis is logarithmic.

Table

ModelQualityLegalityCostn
SolHigh100.0100.0$0.083
LunaHigh55.8100.0<$0.013
Muse Spark 1.3High53.5100.0$0.133
Opus 5.5High45.6100.0$0.043
GLM-5.3 FlashHigh45.666.7$0.023
DeepSeek V4.1 FlashHigh40.4100.0$0.033
Haiku 5.5High22.466.7<$0.013

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

Look