Foaster.ai

Reshaping businesses for the agentic era.

Foaster.ai is a French start-up focused on reshaping businesses for the agentic era. At Foaster Labs, our Werewolf Benchmark studies how LLMs behave under social pressure: leadership, bluffing, and resistance to manipulation.

Results — Podium (role-conditioned Elo)

ELO-W = wolf (manipulation power) · ELO-V = villager (manipulation resistance)

• 🥇 GPT-5 (OpenAI) — ELO 1492 (W 1508 · V 1476), win rate 96.7%, 60 matches
• 🥈 Gemini 2.5 Pro (Google) — ELO 1261 (W 1163 · V 1360), win rate 63.3%, 60 matches
• 🥉 Gemini 2.5 Flash (Google) — ELO 1188 (W 1103 · V 1273), win rate 51.7%, 60 matches