MTG Bench Ranks LLMs: GPT-5.5 Scores 95.4, Deepseek Trails at 12.8

A Magic: The Gathering benchmark scores 15 frontier models on complex game reasoning. GPT-5.5 medium leads at 95.4 for $0.10/query, while gpt-5.4-nano offers the best value at 68.2 score for $0.01. Deepseek-v4-pro scores just 12.8 despite high compute settings, and claude-opus-4-8 underperforms at 39.8.