Skip to content
Hot storyPast story

SpireBench: 20 large models take on Slay the Spire at A20 with just one clear

1 report1 sourceUpdated 2026-10-20

What happened

AI overview

Researchers from Canada's Vector Institute and Radboud University in the Netherlands have posted a preprint on arXiv titled SpireBench, which uses Slay the Spire to evaluate large models' long-horizon strategic ability. All 20 large models played the Ironclad and faced the same 10 new seeds at A20 difficulty; across 200 runs, only GPT-5.6 Sol cleared the game once. Compared with four top streamers, the AI fought only about 0.9 to 1.4 elite battles in Act 1 on average, versus about 2.6 for the streamers; the AI chose to heal at 57% to 75% of campfires, versus about 9% for the experts. The report argues that on the first act alone the best-performing models are already close to top human players, but long-term strategic ability still shows a clear gap, with most losses stemming from overly cautious decisions.

AI-written from the coverage · updated 2026-10-20

Coverage timeline

Follow the coverage to see the story from different sides.

Oct 20
  1. YYS (Youyanshe)Pick
    20 large models take on Slay the Spire at A20 with only one clear, losses mostly blamed on being too cautious

    Researchers used SpireBench to have 20 large models play through the same 10 new seeds at A20 difficulty in Slay the Spire; across 200 runs, only GPT-5.6 Sol cleared the game once. Compared with four top streamers, the AI fought only about 0.9-1.4 elite battles in Act 1 on average, versus about 2.6 for the streamers, and the AI chose to heal at 57%-75% of campfires, versus about 9% for the experts. Counting the first act alone, the best-performing models are already close to top human players, but long-term strategic ability still shows a clear gap.