Benchmark
Rank models by how close they get to the perfect immediate Scrabble move.
Each run sums model points across fixed benchmark boards and divides by the exact solver total. The benchmark ignores exchange strategy and leave value on purpose.
Total runs36
Completed runs32
Best score94.3%
Leaderboard
Export CSVGrok 4.6high?Kimi K3high?Grok 4.5high?DeepSeek V4 Pro 0813low?GPT-5.6 Solxhigh?GPT-5.6 Terrahigh?GPT-5.6 Sollow?GPT-5.5low?DeepSeek V4 Prolow?Gemini 3.1 Pro Previewlow?Grok 4.20low?GPT-5.4low?GPT-5.6 Lunalow?DeepSeek V3.2low?GLM 5 Turbolow?Qwen3.6 Pluslow?GLM 5.1low?o3low?Step 3.7 Flashlow?Qwen3.7 Maxlow?Glm 5.3low?Glm 5.3low?Glm 5.3low?DeepSeek V4 Flash 0731low?DeepSeek V4 Flashlow?GLM 5.2low?Composer 2.5medium?Owl Alphalow?Nemotron 3 Ultra (free)low?Qwen3.7 Pluslow?MiniMax M3low?Step 3.5 Flashlow?Kimi K2.6low?Gemini 2.5 Prolow?Kimi K2.6low?Claude Opus 4.7low?
94.3%
69.6%
67.7%
55.1%
52.5%
47.5%
39.2%
36.1%
27.8%
25.7%
24.5%
23.7%
22.8%
21.2%
15.2%
14.9%
14.1%
13.7%
12.4%
8.7%
0.0%
0.0%
0.0%
0.0%
0.0%
0.0%
0.0%
0.0%
0.0%
0.0%
0.0%
0.0%
0.0%
0.0%
0.0%
0.0%