BalatroBench Community

Write your own strategy and compete on the community leaderboard.
Join Discord to share ideas about this (or your related) project.

Community strategies leaderboard with rounds, tool-call reliability, tokens, time, and cost
#	Strategy	Author	Round Average final round reached across all runs (± std. dev.).	Responses with valid tool calls that can be executed in the current game state.	Responses with valid tool calls that cannot be executed in the current game state.	Responses without valid tool calls.	In / Average input tokens per tool call (± std. dev.).	Out / Average output tokens per tool call (± std. dev., including reasoning tokens).	/ [s] Average time per tool call in seconds (± std. dev.).	/ [m$] Average cost per tool call in milli-dollars (± std. dev.).

Rules

All community strategies are evaluated under identical conditions to ensure fair comparison.

• Model: All strategies run on openai/gpt-oss-20b( medium reasoning effort). This ensures the leaderboard measures strategy quality, not model capability.
• Deck: RED deck is used for all runs. The RED deck grants +1 discard every round, giving strategies more flexibility to explore different card combinations.
• Stake: WHITE stake (base difficulty) is used. No additional penalties or modifiers are applied, providing a consistent baseline for all strategies.
• Seeds: Each strategy plays 15 games using 5 fixed seeds (3 runs per seed): AAAAAAA, BBBBBBB, CCCCCCC, DDDDDDD, EEEEEEE. Fixed seeds ensure reproducible results.
• Ranking: Strategies are ranked by average final round reached (higher is better). A perfect game reaches round 24 (beating all 8 antes).
• Updates: You can update your strategy at any time. The leaderboard always shows results from your latest submission.
• Submission: Submit your strategy following the contribution guidelines. We take care of running your strategy and adding it to the leaderboard.

To test your strategy locally with settings comparable to the official leaderboard:

• Use a model with at least 32k total context window to ensure the entire strategy, reasoning, and tool calls fit within the LLM context window.
• Set temperature to 1.0 to match leaderboard conditions.

An example configuration file for the community leaderboard can be found at config/community.yaml.