Iceburg opens soon. Join the waitlist to get in first. Free, no card.
Measured, not modelled

Benchmarks

Every figure here comes from 3 recorded runs covering 9 prompts, answered by a real AI with and without Iceburg. Nothing is modelled or extrapolated. These are our own test runs, not customer traffic: your savings depend on your prompts, and every response reports its own.

56.7%
Savings (size-weighted)
9,445 → 4,092 tokens
61.1%
Best run
one category
5,353
Tokens saved
across all runs
3
Runs recorded
9 prompts total
DatePromptsWithoutWith IceburgSaved
2026-11-1532,8751,46649.0%
2026-11-1533,5971,40161.1%
2026-11-1532,9731,22558.8%

How to read this

The headline figure is weighted by size: total tokens in versus total tokens billed across every run, so the number reflects real workloads rather than a flattering average.

These are generative prompts (explain, analyse, how-to), each answered by the same AI twice: once as a plain chatbot, once through Iceburg. Without is everything that AI was sent plus everything it wrote; With Iceburg is the same for the Iceburg path. Answers through Iceburg are more concise; a request for a long piece keeps its full length, so that run saves less. Writing costs more than reading at most providers, so the saving on your bill is usually higher than the token count shows.

Token counts use an estimate close to, but not identical to, your provider’s tokeniser. Treat them as a good indicator of the savings, not an exact invoice reconciliation.