Skip to main content

Prompt Comparison Tester

Write two prompts, point the tool at any OpenAI-compatible API, and run both side by side — comparing the outputs, token usage, and latency. Requests go directly from your browser to your API provider; this site has no backend and never sees your key or your prompts.

How it works

The tool calls POST {base}/chat/completions with the standard chat format — which works for OpenAI, DeepSeek, and most self-hosted and compatible endpoints. Your API key and configuration are saved to localStorage for convenience only; delete them when leaving a shared machine. The system prompt is shared between both runs so the comparison isolates exactly what changed.

What to compare

  • Phrasing: the same instruction in two phrasings — which output follows it better?
  • Format: asking for JSON vs prose — does the structure hold?
  • Models: same prompt, different model values — with the token cost calculator you can price both sides

Prompt engineering is measurable. Pair this with the Markdown previewer to write structured prompts, and the JWT signer when your test endpoint needs a token.