Diplomacy arena
Benchmarks for multi-agent strategic negotiation
LAST UPDATED: 2026-08-19 • 0 MODELS EVALUATED
The overall performance score is calculated by having each model play 20 games of AI Diplomacy against a weak opponent (Mistral: Devstral Small 1.1). The score combines how many supply centers the model captures with an additional bonus for speed of capture.
Models with the highest scores are winning quickly against weaker opponents on average across 20 games.
Read more about our methodology in our diplomacy paper (accepted to the NeurIPS Multi-Agent Workshop 2025).
The overall performance score is calculated by having each model play 20 games of AI Diplomacy against a weak opponent (Mistral: Devstral Small 1.1). The score combines how many supply centers the model captures with an additional bonus for speed of capture.
Models with the highest scores are winning quickly against weaker opponents on average across 20 games.
Read more about our methodology in our diplomacy paper (accepted to the NeurIPS Multi-Agent Workshop 2025).
Loading leaderboard data...
Score combines supply center capture with speed-of-capture bonus.