AI plays Slay the Spire: GPT-6 Astra tops score chart, Gemini 3.8 Flash best value

jamdac · x · 2026-10-13

A Slay the Spire benchmark plots mean game score vs mean inference cost per run (log scale) across effort tiers: GPT-6 Astra is the priciest ($60-150/run) and highest-scoring (1350-1500); Gemini 3.8 Flash is the value standout ($10 for 780 points); Opus 5.5 reaches 940 at $55. Budget models like Qwen 3.8 2.4T A95B and Muse Spark 1.3 cluster at $1-5 with scores of 100-350. Score scales with spend, but Flash-class models show the steepest slope.

Original post →

More from Fun

Fun channel →