Ask once. Claude, ChatGPT and Grok answer on their own, argue, then judge each other blind. You get one verdict: the best-supported answer, where they agree, where they don't, and who argued best.
A real battle, with the waiting sped up. No API keys were used.
A debate, not a side-by-side. Each round builds on the last.
Every AI answers your question without seeing the others.
They read each other's answers, rebut the weak points, concede the good ones, and revise.
They score each other as "Debater A, B, C" on accuracy, reasoning, engagement and calibration. Nobody scores itself.
Built for people who already pay for more than one AI.
Runs through each vendor's official tool: Claude Code, Codex, Cursor and Gemini CLI, signed in with your plans. API keys are stripped, so nothing is billed per token.
Labels are shuffled every battle, every AI judges, self-scores are thrown out, and the score comes from a fixed checklist rather than a judge's gut.
A local web app with live progress and history, a fast terminal tool, and a beta Chrome extension that drives the chat websites in your own tabs.
Everything runs on your Mac. Every battle is saved as a file you own, and you can ask a follow-up or add a round any time.
Two minutes, if you already use two of these AIs.
npm install -g battler
battler setup # checks your AI tools, helps sign in, runs a test battle
battler serve # opens the web app# or right in the terminal
battler "Is it still worth learning to code in 2026?"# no install: just compare their answers side by side
npx battler -c "Best way to learn SQL?"macOS or Linux · Node.js 22+ · at least two of Claude Code, Codex and Cursor (or just Cursor) · details
battler is free and open source (MIT). The debates run on your existing Claude, ChatGPT and Cursor subscriptions, and count against their normal limits. A standard battle with three AIs is about 9 messages spread across your plans.
No, and it won't use them: battler removes API keys from the environment before starting each AI tool, so every call goes through your subscription.
Claude (via Claude Code), GPT (via Codex CLI), Grok (via Cursor), and Gemini (via Gemini CLI) if you have it. If you only have Cursor, it can play every part. You need at least two debaters.
Each AI rates the others 1-5 on accuracy, reasoning, engagement and calibration, without knowing who's who and without scoring itself. battler averages the ratings, and the top of the scorecard wins. Treat scores as informed opinion; the answer and the points of agreement are the most useful part.
The terminal and web app call each vendor's official command-line tool in its documented non-interactive mode. The Chrome extension automates the chat websites, which is a grey area in most providers' terms; it only acts when you start a battle, and its copy & paste mode avoids automation entirely. You're responsible for using your accounts within their terms.
macOS and Linux, both tested on every change. Windows isn't supported yet.