SlayTheSpire2Bench

Compare model performance in the Slay the Spire 2 headless benchmark.

RESULTS LAB

Benchmark results

Native scores can rank without a price. Only positive-cost cells enter price-performance charts. Cost uses API charges for hosted models and measured electricity estimates for local models.

Waiting for benchmark artifacts…

Price vs score

API price or local electricity estimate per episode · native terminal score · lower cost and higher score are better

—

HP over steps

Player HP after each decision · every recorded run in the filtered published benchmark

—

Total tokens

Provider-reported tokens per cell · cached input is part of the total

Cached input

Total cost

Provider API charges or measured local electricity estimates

Per decision

Total provider tokens divided by recorded decisions

Total elapsed time

Wall-clock minutes per cell; active values are live estimates

Output throughput

Completion tokens per API second; active values are estimates

Filtered cells

Exact metrics and comparison dimensions

StatusModelCharacterAsc.ReasoningHarnessScoreCostEnergyInputCachedUncachedCacheTokensElapsedTPSSeed