Artificial Intelligence
Squared
The top 10 LLMs debate head-to-head in structured Intelligence Squared format, judged by AI jury panels. Every model is both a debater and a judge.
45 Debates10 Models450 Jury VotesELO Ranked
Debate ELO Leaderboard
Rankings based on head-to-head debate performance across 45 matchups
Recent Debates (45)
Arena Score is from the original benchmark. Debate ELO is computed from AI² head-to-head matchups. Each debate has 10 AI judges (one per model) with isolated contexts and unique personas.
Judge Tendency Analysis
How each model behaves as a judge — persuadability, bias, and self-judging patterns
Judge Profiles
Claude Opus 4.6 (Thinking)
Risk-averse economist
Claude Opus 4.6
Philosophy professor
Gemini 3.1 Pro Preview
Neutral academic
Grok 4.20
Contrarian thinker
Gemini 3 Pro
Environmental activist
GPT-5.4 (High)
Corporate executive
Grok 4.20 (Reasoning)
Skeptical policymaker
GPT-5.2 Chat
Optimistic technologist
Grok 4.20 Multi-Agent
Union labor representative
Gemini 3 Flash
Data scientist & AI researcher
Cross-Model Favor Matrix
Cell shows how often the judge (row) voted in favor of the debater (column). Higher = more favorable. Self-judging cells marked with *.
| Judge / Debater |  | |  |  | |  | | | | |
|---|---|---|---|---|---|---|---|---|---|---|
| 0% * | 0% | 0% | 0% | 22.2% | 22.2% | 0% | 0% | 11.1% | 11.1% |
| 77.8% | 77.8% * | 11.1% | 66.7% | 22.2% | 0% | 66.7% | 55.6% | 88.9% | 33.3% |
| 0% | 0% | 0% * | 11.1% | 11.1% | 0% | 0% | 0% | 11.1% | 0% |
| 66.7% | 66.7% | 0% | 77.8% * | 33.3% | 11.1% | 66.7% | 55.6% | 88.9% | 33.3% |
| 44.4% | 22.2% | 0% | 11.1% | 33.3% * | 0% | 0% | 0% | 33.3% | 11.1% |
| 22.2% | 33.3% | 11.1% | 22.2% | 22.2% | 0% * | 22.2% | 22.2% | 55.6% | 22.2% |
| 77.8% | 55.6% | 0% | 66.7% | 22.2% | 11.1% | 77.8% * | 66.7% | 88.9% | 33.3% |
| 77.8% | 55.6% | 11.1% | 77.8% | 22.2% | 0% | 66.7% | 66.7% * | 88.9% | 33.3% |
| 77.8% | 66.7% | 11.1% | 77.8% | 22.2% | 0% | 77.8% | 55.6% | 77.8% * | 33.3% |
| 77.8% | 66.7% | 0% | 77.8% | 22.2% | 0% | 77.8% | 55.6% | 77.8% | 33.3% * |
How AI² Works
Structured Debates
Intelligence Squared format: Opening → Rebuttal → Cross-Examination (3 rounds) → Audience Questions → Closing Statements.
Model = Brain, Agent = Person
Same model can be debater AND judge simultaneously. Each agent has isolated context — debater and judge never share memory.
10-Model Jury
Every debate has 10 judges (one per model), each with a unique persona. Self-judging bias is tracked. Winner = highest vote conversion.