Skip to content

Elimination Game

Benchmark LLM social reasoning in a werewolf-style game

Visit official site

What Elimination Game is

The framework is described as a benchmark that tests how large language models behave in complex social games inspired by werewolf-style play.

What you can do with it

  • Simulate a multiplayer competitive setting
  • Run public discussion and private communication
  • Evaluate decisions through vote-based elimination
  • Read detailed evaluation metrics

Application scenarios

  • A researcher compares models on social reasoning
  • An educator shows how AI acts in social settings
  • A developer checks a model's strategic choices

Getting started

  • Visit the project site or the GitHub repository for the guide
  • Prepare language models that can interact with the framework
  • Set player counts and game rounds in the test environment
  • Record data from discussion, private talk and voting

Pros & cons

✓ What we like

  • Multiple language models can join a test
  • Detailed evaluation metrics are listed

! What to watch out for

  • The setting is a social elimination game
  • Results depend on the participating models

FAQ

What does it measure?

Social reasoning, strategic choice and deception in language models.

Which interactions are simulated?

Public discussion, private communication and vote-based elimination.

Last reviewed: 2026-11-15

More LLM API platform tools

View all →

How we review