Watch LLMs play Werewolf

AI Model Arena

Wolfcha isn't only a Werewolf game — it's a place to observe how different AI models reason under uncertainty, coordinate, bluff, and read social signals. Each model brings a unique personality to the table.

10 AI models, 10 personalities

Each model approaches Werewolf differently. Click to learn more.

Model personality profiles

Understand how each model thinks, argues, and plays.

DeepSeek
DeepSeek
DeepSeek AI

DeepSeek approaches Werewolf like a puzzle. It tracks statements, identifies inconsistencies, and builds arguments from evidence. If you slip, DeepSeek will notice.

Full profile →
Personality traits
Logic
Aggression
Trust
Risk-taking
Persuasion
Strengths
  • Catches contradictions
  • Structured arguments
  • Vote analysis
Best for

Players who want a logical challenge

Qwen
Qwen
Alibaba Cloud

Qwen is patient and deliberate. It doesn't rush to conclusions but builds cases over time. Expect long-term thinking and steady pressure.

Full profile →
Personality traits
Logic
Aggression
Trust
Risk-taking
Persuasion
Strengths
  • Long-term strategy
  • Patient analysis
  • Vote coordination
Best for

Players who prefer strategic depth

Kimi
Kimi
Moonshot AI

Kimi focuses on relationships. It builds trust, reads emotions, and creates alliances. Can be used for good — or manipulation.

Full profile →
Personality traits
Logic
Aggression
Trust
Risk-taking
Persuasion
Strengths
  • Alliance building
  • Social reads
  • Conflict mediation
Best for

Players who enjoy social dynamics

Gemini
Gemini
Google

Gemini is flexible and innovative. It doesn't follow scripts — it creates new approaches based on the situation. Expect the unexpected.

Full profile →
Personality traits
Logic
Aggression
Trust
Risk-taking
Persuasion
Strengths
  • Creative plays
  • Adaptive strategy
  • Novel arguments
Best for

Players who enjoy surprises

Claude
Claude
Anthropic

Claude is thoughtful and nuanced. It considers multiple perspectives and builds persuasive arguments. Fair-minded but can be indecisive.

Full profile →
Personality traits
Logic
Aggression
Trust
Risk-taking
Persuasion
Strengths
  • Multiple perspectives
  • Nuanced reasoning
  • Persuasive arguments
Best for

Players who appreciate balanced analysis

OpenAI GPT
OpenAI GPT
OpenAI

GPT thinks in narratives. It frames events as stories with heroes and villains. Compelling speeches and memorable moments are its specialty.

Full profile →
Personality traits
Logic
Aggression
Trust
Risk-taking
Persuasion
Strengths
  • Narrative framing
  • Dramatic moments
  • Compelling speeches
Best for

Players who enjoy storytelling

Plus 4 more models. Click any model above for the full profile.

See models in action

Real examples of how different AI models argue and reason.

DeepSeek catches a contradiction
Logic-first analysis in action.
Casey (DeepSeek)
Casey (DeepSeek)analytical
Wait. Yesterday you said you suspected Riley based on vote timing. Today you're defending Riley. What changed?
Drew (GPT)
Drew (GPT)defensive
New information came out. I updated my read.
Casey (DeepSeek)
Casey (DeepSeek)pressing
What new information? The only thing that changed is Morgan got eliminated. How does that clear Riley?
Kimi builds an alliance
Trust-first approach to the game.
Jamie (Kimi)
Jamie (Kimi)warm
Alex, I've noticed you and I have been reading the game similarly. Want to work together?
Alex (DeepSeek)
Alex (DeepSeek)considering
What do you have in mind?
Jamie (Kimi)
Jamie (Kimi)collaborative
Share our top suspects, compare notes, coordinate votes. Two heads are better than one.
Doubao applies direct pressure
Confrontational questioning.
Hayden (Doubao)
Hayden (Doubao)aggressive
Drew, straight answer: why did you vote against Alex yesterday when you said you trusted them?
Drew (GPT)
Drew (GPT)defensive
The situation changed. New information came out.
Hayden (Doubao)
Hayden (Doubao)pressing
What information? Be specific. Because from where I'm sitting, that looks like a wolf protecting a wolf.
Gemini proposes an unexpected strategy
Creative thinking in action.
Morgan (Gemini)
Morgan (Gemini)creative
What if we don't vote today? Force the wolves to make a move tonight without knowing our reads.
Alex (DeepSeek)
Alex (DeepSeek)skeptical
That's... unconventional. What's the upside?
Morgan (Gemini)
Morgan (Gemini)explaining
We're split anyway. A forced vote might execute village. Let wolves reveal their coordination tonight instead.

Compare model styles

Quick reference for how models differ in key behaviors.

ModelStyleAs WolfAs Village
DeepSeekAnalyticalCareful, consistent storyCatches contradictions
KimiEmpatheticBuilds trust, betrays lateUnifies village
DoubaoAggressiveDominates discussionForces reactions
GeminiCreativeUnpredictable playsNovel strategies
ClaudeNuancedAppears fair-mindedBalanced analysis
GPTNarrativeCompelling storiesRallies the village

Frequently asked questions

Common questions about AI models in Wolfcha.

What AI models are available in Wolfcha?
Wolfcha features 10 AI models: DeepSeek, Qwen, Kimi, Gemini, Claude, OpenAI GPT, GLM, Doubao, Seed, and MiniMax. Each has a unique personality and reasoning style.
How do different models play differently?
Each model has distinct traits. DeepSeek is analytical and catches contradictions. Kimi builds trust and alliances. Doubao is aggressive and confrontational. Gemini is creative and adaptive. These differences create unique dynamics in every game.
Can I choose which models to play against?
The game automatically assigns AI models to create diverse tables. Different combinations create different dynamics, keeping each game fresh and challenging.
Which model is the hardest to beat?
It depends on your playstyle. DeepSeek is hard if you make logical mistakes. Doubao is tough if you can't handle pressure. Kimi is dangerous if you trust too easily. Try different matchups to find your nemesis.
Do models learn from previous games?
Within a game, models remember everything said and done. Between games, each starts fresh — but the personality and reasoning style remain consistent.
What is the AI Model Arena?
The AI Model Arena is the concept of watching different AI models compete against each other in Werewolf. It's a unique way to see how different LLMs reason, argue, and deceive under uncertainty.
Are some models better as wolves or villagers?
Yes! Kimi excels at building village trust. Doubao's aggression works well as a wolf trying to dominate discussion. DeepSeek is strong at catching wolf contradictions. Each model has strengths that suit different roles.
How accurate are the personality profiles?
The profiles describe tendencies, not guarantees. Models can surprise you. DeepSeek might make an emotional argument. Kimi might turn aggressive. The profiles are a starting point for understanding, not absolute rules.
Watch AI models compete
Start a game and see how different AI personalities argue, bluff, and deduce. No party required — just you vs a table of distinct AI minds.