Ai Systems Are Improving Rapidly At Slay The Spire And Its Sequel
Source
epoch.ai
Author
Epoch AI publications
Date
Why it matters
Documents rapid gains in long-horizon game play via computer use, a proxy for sequential decision-making by agents, along with caveats about remaining elicitation gaps.
Key takeaways · AI-distilled
At Slay the Spire 1's maximum difficulty, Ascension 20 Heart, GPT-6.1 Sol wins 20% of runs and GPT-6 Astra 17%, while top human players regularly win at least 80% at maximum difficulty in both games.
The jump is recent: in June 2025 the Orak paper found Gemini 2.5 Pro reached the second boss in two of three runs, likely a 0% win rate, and in June 2026 Anthropic said Fable 5 reached the final act three times more often than Opus 4.8, short of reliable wins.
Frontier models top 25% at Slay the Spire 2's hardest setting, Ascension 10, better than at comparable tiers of the original, which Epoch attributes to the sequel's lower overall difficulty in early access.
Epoch calls earlier attempts to benchmarkA standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.Full definition → Slay the Spire with LLMs scattershot, with harnesses, allowed affordances and tested models varying substantially between them.
Terms in this piece · Glossary
benchmark — A standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.
