Token counting
When the server proxy is configured, the counter calls Anthropic count_tokens with model: "claude-fable-5". If the API is unavailable, the UI keeps working in estimate mode and labels the result.
Measure tokens, meter cost, and audit every cache tier before a prompt ships. Three calibrated instruments. Claude pricing only. No signup.
Specimen in, calibrated token magnitude and per-call cost out.
Open counter 02 CH-02 EstimateMeter input, output, batch, and every cache tier down to the line item.
Open calculator 03 CH-03 LookupCheck pricing, access, June 22 changes, fallback behavior, and migration gotchas in one place.
View specsNo non-Claude providers are exposed in this MVP. The code structure can expand later when search data and verified pricing justify it.
| Claude model | Input | Output | Cache read | Batch input/output |
|---|---|---|---|---|
| Claude Fable 5 | $10.00 / MTok | $50.00 / MTok | $1.00 / MTok | $5.00 / $25.00 |
| Claude Opus 4.8 | $5.00 / MTok | $25.00 / MTok | $0.50 / MTok | $2.50 / $12.50 |
| Claude Sonnet 4.6 | $3.00 / MTok | $15.00 / MTok | $0.30 / MTok | $1.50 / $7.50 |
| Claude Haiku 4.5 | $1.00 / MTok | $5.00 / MTok | $0.10 / MTok | $0.50 / $2.50 |
When the server proxy is configured, the counter calls Anthropic count_tokens with model: "claude-fable-5". If the API is unavailable, the UI keeps working in estimate mode and labels the result.
Regular input, cache writes, cache reads, and output are priced separately. Batch and US-geo modifiers are applied after the base cost, matching the current MVP contract.
The MVP proxy does not store pasted text. It only needs request status, rough character bucket, and timing for abuse and reliability checks.
Pricing changes should be fixed in PRICING_CONFIG, rebuilt, and noted with a new last-verified date.
Pricing constants are isolated in one config file and should be rechecked before launch.
Estimates are for planning. Check your Anthropic invoice and current pricing before committing spend.