Prompt Comparison / A-B Analyzer
Compare prompt tokens, input costs, context usage, variables and line differences locally.
16 KiB UTF-8 per prompt. Text stays inert; neither prompt runs. Current catalog tokenizers provide estimates only. Characters are code points; words are whitespace-delimited. Costs are Standard uncached text input only, per request. Context remaining is before output, wrappers or other reserves.
Line diff
Compare to see changes.
Ready. Compare text without executing it.
Model context source · 2026-10-08. Input pricing source · Effective 2026-10-08, reviewed 2026-10-08. No caching, output, tools, taxes or account discounts included.
Enable JavaScript and Web Workers to compare locally.
Compare before running an experiment.
Inspect two prompt versions without executing either. Differences describe size and structure; they do not choose a winner or predict response quality. Each prompt allows 16 KiB of valid UTF-8 text.
Estimates, not model execution.
Tokens reuse the Token Counter heuristic. The current catalog provides estimates only, not exact vocabulary tokenization. Input costs use the selected model and pricing date; missing pricing stays unavailable. Remaining context excludes output reserves, message framing and tools.
Literal variables and structure.
Double-brace ASCII identifiers use the Prompt Template Playground parser. Shared and exclusive names show parity; repeated occurrences are counted separately. Headings, bullets, numbered lines, fence delimiters and XML-like tags are literal signals, including inside code blocks, not semantic analysis.
Bounded line differences.
The shared Text / Code Diff worker compares lines, treating CRLF and LF alike. Modified lines appear as removed and added lines. Large differences may hit the existing line, edit or output limits while other metrics remain available. Inputs stay in memory and clear when leaving this page.