Paste a transcript. See your input bill drop up to 75%.

Drop in a real conversation and this tool does the math your bill would do: how many input tokens PromptCrunch forwards, how many it strips, and what you pocket every month. Identical responses, a fraction of the input cost.

Runs entirely in your browser. Your transcript never leaves this page.
Paste
A real conversation
Simulate
Exactly what we'd forward
Read
Your monthly savings

Already using prompt caching (cache_control)?

Good, keep it. Cached requests pass through untouched, so PromptCrunch is never worse than sending direct, and it crushes the input bill on every request that isn't a cache hit. See how caching and PromptCrunch stack.

The rules behind your number

These are the live proxy's real thresholds, the same math your production traffic runs through. Read them and check our work.

  1. We count tokens fast. Characters ÷ 4 lands within about 10% of a real tokenizer on English prose. Base64 image blocks count as a flat ~1,500 tokens each.
  2. Short chats ride free. Anything at or under 12,000 tokens is forwarded untouched, so you're never charged more than going direct. Compression kicks in exactly where the money is: long, growing conversations.
  3. The latest turns stay whole. The last 4 messages always go through verbatim, so the model keeps full fidelity on what was just said.
  4. Your code is safe. Fenced code blocks, JSON-looking messages (starting with { or [), base64 runs, images, tool calls, and tool results in older history pass through verbatim. Never compressed, never touched.
  5. The prose gets trimmed. Older conversational text is compressed toward a 0.4 ratio: forwarded = protected + 0.4 × compressible + the last 4 messages.
  6. No inflated numbers. These are the exact rules the live proxy runs. The win you see here is the win you'd get in production - not a demo number.

Now run it on your own traffic.

Two lines of config point your SDK at PromptCrunch. The free tier hands you $5 of credit and 100 requests a day, no card, so your real conversations prove the number for themselves.

Start free See where caching stops and we start