Tokens Boutique

Here we are with version v26.0 and a huge update! A lot of new features and fixes.

FEATNew Prompt Analyzer. Paste any prompt and it shows you a token-by-token breakdown, an estimated token count, and the cheapest models for that exact prompt across every vendor. Everything runs locally in your browser, nothing is ever sent anywhere.

For convinience you can also access it via analyze.tokens.boutique!

FEATHover any token to see its exact text and byte size, and whether it was flagged as part of a repeated or redundant section.

FEATThe analyzer also flags redundant or repeated text in your prompt, with an estimate of how many tokens you could save by trimming it.

FEATIt then ranks every model by estimated cost for that specific prompt, cheapest first.

FEATYou can also tell it how long you expect the reply to be: assume a similarly sized reply (the default), a model's own maximum output, or a specific number of tokens.

FEATAdded cache-hit pricing. Many vendors charge a lower rate for input tokens that get served from a prompt cache. Flip the new toggle and every price in the table (and in the analyzer) updates to reflect that cheaper rate.

FEATNot every vendor publishes cache-hit pricing for every model. When that's the case, we show a small orange marker and fall back to the regular price instead of guessing.

FEATHovering the deprecated marker (and now the cache-hit marker) shows a proper tooltip instead of relying on your browser's default hover text.

FEATYour Frontier only, No deprecated, and Cache-hit prices choices on the main comparison table are now remembered between visits.

FIXFixed a bug in the Mistral pricing table where two current models were incorrectly marked as deprecated.

FIXFixed a visual bug where the blinking cursor in the Prompt Analyzer's text box could drift out of sync with the highlighted text while typing.

FIXFixed a bug where a tooltip could render partially outside the page on narrower screens.