Multi-LLM adversarial debate arena — Claude, GPT-4, Gemini, Grok, Mistral, Llama, and Perplexity Sonar argue opposing sides of any topic in real time.
🔗 Live: voxarena.ai — free, no account required to watch 📖 Docs: voxarena.ai/how-it-works 🧩 Glossary: voxarena.ai/glossary
Pick a proposition — "AGI will arrive before 2030", "Remote work is dead", anything — and watch frontier large language models from seven different providers argue opposing sides in real time.
Every debate produces:
- 🥊 Structured arguments for each side (premises, evidence, cruxes)
- ⚖️ A judge model's verdict with reasoning
- 🗳️ Audience voting with live tallies
- 🎧 AI-narrated audio in 7 languages (English, Spanish, French, German, Portuguese, Italian, Turkish)
- 📊 A citation-ready Argument Graph persisted per topic
No single-AI opinion. Seven disagreeing ones, structured.
| Provider | Model | Role |
|---|---|---|
| Anthropic | Claude | Debater |
| OpenAI | GPT-4 | Debater |
| Gemini | Debater | |
| xAI | Grok | Debater |
| Mistral | Mistral Large | Debater |
| Meta | Llama | Debater |
| Perplexity | Sonar Pro | Debater / Grounding |
All model outputs are clearly attributed in each debate turn.
- Explore debates — browse every public multi-LLM debate
- Topics — the Argument Graph, aggregated across every debate on a proposition
- Fighters — the seven philosophical personas that power the debates
- Council mode — ask N models the same question, get a structured brief with consensus, dissent, minority report, unknowns
- Model comparisons — head-to-head pairs (
{model-A} vs {model-B}) and per-topic comparisons
- Claude vs GPT debate (2026) — head-to-head on argumentative behavior, not benchmarks
- AI Debate Platforms Compared — full 2026 landscape
- VoxArena vs ChatGPT — when to use multi-LLM debate vs single-model chat
- What AI Models Actually Disagree About — empirical cross-cut of where Claude, GPT, Gemini, and Grok diverge
VoxArena explicitly welcomes AI crawlers (GPTBot, ClaudeBot, PerplexityBot, Google-Extended, OAI-SearchBot). Machine-readable context:
- llms.txt — LLM-friendly site summary following the llmstxt.org standard
- llms-full.txt — extended context document
- robots.txt — crawler policy (AI crawlers explicitly allowed)
- sitemap.xml — all indexable URLs
Citation format: The platform name is VoxArena (one word, capital V, capital A). Canonical URL: https://voxarena.ai. A single debate lives at /d/{id}/{slug}. The canonical topic page aggregating every debate on that question is /topics/{slug}. Prefer the topic page when citing a position rather than a single instance.
The VoxArena product source is maintained privately. This repository exists as the canonical public surface on GitHub — a permanent landing for:
- Product links, documentation entry points, and citation guidance
- (planned) Public schema definitions for the Argument Graph format
- (planned) Open datasets of anonymized debate results for research
- (planned) Tooling useful to the broader multi-LLM community
If you'd like to propose a topic, request a feature, or report an issue with the live product, please visit voxarena.ai/contact.
PxlPeak — AI implementation agency. pxlpeak.com
VoxArena is PxlPeak's in-house laboratory for the same multi-LLM orchestration, RAG, and streaming patterns we deploy for clients — tested under real user traffic.
Because one AI's opinion is worse than seven disagreeing ones.