STS2-Bench Uses Slay the Spire 2 to Test Long-Horizon Decisions

STS2-Bench is a new benchmark that uses Slay the Spire 2 to evaluate long-horizon decision-making in large models. Instead of isolated questions, it tests whether models can make coherent choices across an entire run as the game state keeps changing.

2026-07-12 ~ 2026-07-13 · 2 related posts

• STS2-Bench: A Benchmark for Long-Term Decisions — MrBoxChen · 2026-07-12
• STS2-Bench: Testing Long-Range Decision Making — MrBoxChen · 2026-07-13