Light questions go to the fast, cheap model; only complex tasks use the premium one. It gets cheaper automatically — no thinking required.
COST TELEMETRYReal cost, per message
See the actual cost of every answer, plus today/month totals and the amount saved by optimization. Auto-displays in 166 currencies.
ZERO TRUST NEEDEDYour key and chats never leave this device
No account. No server. Your API key and conversations are stored only in your browser; the app talks only to Anthropic.
SMART COMPRESSIONLong chats stay cheap
Automatic history compression and prompt caching keep costs down even as conversations grow.
THE 90% TRUTH
88% of questions: the cheap model was practically on par with the top model (measured: 7 of 8 typical questions).
~1/7 the cost — about one-seventh of the top model, for the same job done (measured).
"Cheaper means worse" turned out to be a myth. In our tests, most everyday questions show no felt difference (the top model still wins on some complex judgment or translation).
And you can verify it on your own questions. Hit "Audit" on any answer to send the same question to the top model, judge whether they're on par, and log it as your quality-match rate. Don't take our word for it — prove it.
PRICING
Option
Typical monthly cost
Subscription AI chat
$20–$130
Fixed fee even in idle months
TOKENSE (your own API key)
~$0.0002–0.01 / question
Pay per use. Light users spend well under $1/month
* Measured (Haiku, short questions). TOKENSE itself is free — you pay only Anthropic for API usage.
GET STARTED
Create an API key at console.anthropic.com (free sign-up; a $5 top-up covers thousands of questions)
Open TOKENSE and paste the key (stored only on your device)
Just chat. Costs are measured and optimized automatically
FAQ
Is it safe to enter my API key?
We never receive it. The key lives only in your browser (localStorage) and is never sent to our servers — we don't even have one. The app talks only to Anthropic's API, plus one daily FX-rate fetch.
How is this different from Claude Pro/Max?
It coexists rather than replaces. Use it as the place to continue when you hit your Pro/Max limit, or for light questions at pay-per-use prices. A reset-countdown timer is built in.
Does it work on mobile?
Yes — it runs in any browser, and you can add it to your home screen as a PWA.
What does "SAVED" mean?
The cumulative difference between the theoretical cost of running your questions unoptimized (premium model, no caching) and what you actually paid. It measures how much TOKENSE's optimization is working for you.