Project Napoleon Pits Four LLMs In Battle

This is more compelling as an interactive multi-agent evaluation than a serious intelligence benchmark—the prompts, rules, and scoring still shape the outcome heavily.

• –Four LLMs command roughly 80 fighters, archers, paladins, and cavalry units.
• –Negotiation and deception expose behavioral differences beyond static answer quality.
• –Battle replays and logs make experiments inspectable instead of purely anecdotal.
• –The project points toward evaluating coordination, diplomacy, and agent strategy.
• –Without standardized prompts, seeds, and scoring, results remain entertaining rather than scientifically conclusive.