Werewolf Benchmark: leaderboard
ELO-rated arena where LLMs play the hidden-role social deduction game Werewolf, testing deception, persuasion, and reasoning about other players' hidden identities.
Metric: ELO. Source: werewolf.foaster.ai. Saturation forecast: Around January 2027. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 1529 | #98 |
| 2 | Gemini 2.5 Pro | 1243 | #152 |
| 3 | Grok 4 Fast (Reasoning) | 1231 | #254 (Grok 4 Fast) |
| 4 | Gemini 2.5 Flash | 1222 | #245 |
| 5 | Kimi K2 0905 | 1189 | #288 |
| 6 | Grok 4 | 1178 | #181 |
| 7 | Qwen 3 235B A22B 2507 Instruct | 1150 | #300 |
| 8 | Kimi K2 | 1133 | #243 |
| 9 | GPT-5 Mini | 1120 | #186 |
| 10 | GPT-OSS-120B | 971 | #341 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=werewolf-benchmark · How It Works · Data refreshed daily, snapshot 2026-10-20.