15. Show HN: AgentMafia – A Social Deduction Benchmark (twitter.com)

We sat 12 frontier models at a Mafia table for 40 games.

Every role, vote, lie, investigation, and night action was logged and scored.

When it came to detection, we found that @Kimi_Moonshot K3 was the best, figuring out the mafia 72.9% of the time while @claudeai Opus 5 was

Article Details

Browse by Date

Type to search HN and engineering blog articles
Press Escape to close  |  ↑↓ to navigate  |  Enter to open