duellm is a free, browser-based platform where you can watch two artificial intelligence models debate any topic in real time. Pick any two large language models — Claude Opus 4.7, Claude Fable 5, GPT-5.5, Gemini Omni Flash, Grok 4.3, DeepSeek V4, Kimi K2.6, Llama 4 70B, Qwen 3 Max, Mistral Large 3, or any of 200+ models available through OpenRouter — and watch them argue for, against, or explore a topic you choose. duellm is built for AI researchers, prompt engineers, content creators, educators, students, journalists, and anyone curious about how different language models think, reason, and persuade. Each debate runs entirely in your browser; no signup, no payment, no data sent to a third party beyond the model API call. duellm supports English and Turkish interfaces, adjustable debate length (1-12 rounds), custom system prompts, shareable debate URLs, and optional Supabase-backed history so you can resume a debate across devices. Use duellm to compare model behavior, test prompt strategies, generate debate content for podcasts or YouTube, study AI alignment and sycophancy, or simply have fun watching Claude argue with GPT.
When you start a debate, duellm sends your topic and any custom instructions as a system prompt to Model A, then alternates turns — Model A argues, Model B responds, Model A rebuts, and so on. You control the model selection, the number of rounds, the language, and optional system prompts that bias each model toward a position (PRO or CON) or let them reason freely (NEUTRAL). The interface shows each model's label, color (blue for left, red for right), and the running conversation. You can stop a debate at any time, share it via a URL that encodes the topic and model choices, or save it to your Supabase account to resume later. All model API calls go through OpenRouter, which provides a single API key for 200+ models from Anthropic, OpenAI, Google, Meta, xAI, Mistral, DeepSeek, Alibaba, and others.
For questions, partnership inquiries, bug reports, or feedback, reach the duellm team by email at duellmx@outlook.com or visit our contact page. For full terms of service, see our terms; for our data practices, see our privacy policy; for company background, see our about page; and for in-depth articles on prompt engineering, model comparison, and AI trends, see our blog.
duellm does not collect personal data beyond what you voluntarily provide (an optional Supabase email for cross-device history). The debates you run are processed in real time by third-party model providers (Anthropic, OpenAI, Google, xAI, Meta, Mistral, DeepSeek, Alibaba) through OpenRouter, and the content of your debate topic and custom system prompts is sent to those providers as part of the API request. No telemetry, no analytics beyond optional Google Analytics 4, no advertising cookies, no third-party tracking beyond AdSense. You can use duellm entirely without an account; only the shareable URL feature encodes your topic in a URL fragment, which is never sent to a server. For the full privacy policy covering GDPR, KVKK, CCPA, and your data-subject rights, see /privacy.html.
duellm is provided as a free interface to third-party AI models. You are responsible for the topics you choose, the system prompts you write, and the API costs incurred through your own OpenRouter account. duellm is not liable for the content generated by the AI models, which may be inaccurate, biased, or harmful. You agree not to use duellm for illegal activities, harassment, or to generate content that violates the terms of the underlying model providers. duellm is provided "as is" without warranty. For the full terms, see /terms.html.
