LudoBench: leaderboard

Strategic decision-making in the board game Ludo on LudoBench (2026): 480 handcrafted single-decision positions in 12 categories, each isolating one tradeoff such as capturing, reaching safety or finishing a piece, played with a neutral persona. Share of positions where the model moves the same piece as a depth-limited expectiminimax agent, in percent; higher is better.

Metric: Game-theory alignment (%): share of the 480 LudoBench board positions (12 decision categories, neutral persona, legal moves withheld, temperature 0) where the model moves the same piece as a depth-limited expectiminimax game-theory agent; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 6 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Gemma 3 12B (IT)46#1907
2Qwen 2.5 7B Instruct46#2732
3DeepSeek-Chat (LudoBench checkpoint unspecified)46
4Claude 3.5 Haiku45#1597
5Qwen Plus (LudoBench checkpoint unspecified)43
6Llama 4 Scout40#1825

No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=ludobench · How It Works · Data refreshed daily, snapshot 2026-11-22.