Feed
researchAI Research6d ago

Researchers develop zero-search reinforcement learning agent for Slay the Spire 2

Facts

Researchers have released an AI agent for the game Slay the Spire 2 that utilizes a 6.6M-parameter entity transformer to make decisions without lookahead. The agent employs a failure-driven curriculum and restart strategy to achieve performance levels competitive with existing tree-search methods at lower difficulty settings.

  • ▸The AI agent uses a 6.6M-parameter entity transformer architecture.
  • ▸The agent operates without lookahead search, requiring 21-27 ms per decision on a single CPU core.
  • ▸At Ascension 0 difficulty, the model achieved an 80.2% win rate, comparable to the tree-search-based Spire Mentor.
  • ▸The training method utilizes a failure-driven curriculum that allocates restarts based on where games are lost.
  • ▸Code and trained policies have been released on GitHub.
Occurred
2026-11-09 09:05 UTC
First reported
2026-11-09 09:05 UTC
Last updated
2026-11-10 07:17 UTC
Event type
research
Region
global

Entities

Categories

Source timeline · 1

  • arXivFirst-party2026-11-09 09:05
Community vote

Sign in to vote on importance and sentiment.

Platform Signals

Impact score
60
AI sentiment
Positive
Event significance30%0
Entity importance20%0
Source credibility20%1
Novelty15%1
Corroboration15%1
Confidence1
Novelty1
Corroboration1
Community
ImportanceNo votes
Strategic sentiment
Positive 0%Neutral 0%Negative 0%

Platform signals are algorithmic and community-derived indicators. They are not investment recommendations and do not predict valuation or performance.