LLM Output Diff
Side-by-side diff of two model outputs with word-level highlighting and token counts per side.
It highlights exactly what changed between two AI answers.
What this tool does
Comparing two model outputs by eye is unreliable: the answers are long and mostly similar. This tool diffs two pasted outputs at the word level, highlights insertions and deletions side by side, and shows length and token estimates for each version.
- Compare outputs from two models on the same prompt
- See exactly what a prompt change did to the answer
- Check a regenerated answer against the previous one
Output A
chars · words · ~ tokens
Output B
chars · words · ~ tokens
Token estimates use a chars÷4 heuristic. Real tokenizers vary by model and content type.
·
Paste both outputs to compare similarity.
Diff truncated to the first 3,000 words per side for performance. Stats above use the full text.
Paste both model outputs above to see a word-level diff with highlighting.
Your outputs never leave the browser. Diffing, stats, and highlighting run entirely on your device. Nothing is sent to a server.
Comparing two LLM answers by eye is slow when phrasing shifts but the meaning stays close. This tool tokenizes each output into words, finds the longest common subsequence, and highlights insertions and deletions at word granularity so you can spot rewrites fast.
Similarity is 2 × shared words ÷ (words in A + words in B), derived from the LCS length. Use inline mode for one merged view, or side by side to read each answer in context. Pair it with the token cost calculator when you are trimming verbose outputs.