Tokens Boutique

The more vendors we add, the more it's become clear that a lot of real-world pricing doesn't fit cleanly into "$X per million input tokens, $Y per million output tokens."

This release v26.1 is about showing you the facts when calculating estimated costs.

FEATOn the site and in the analyzer, any price that isn't a clean per-token rate now gets a tag next to it. Hover it for the full cost breakdown.

FEATPerplexity charges a flat per-request search fee on top of input/output tokens, and the fee itself has three tiers depending on how much search context is used. The sky tag shows that the medium tier (API default) is already included in est. cost, plus the low/high figures and the opt-in Pro Search rate.

FEATsonar-deep-research goes further and bills separate citation, reasoning, and search-query token charges that can't be derived from input/output tokens alone. The blue tag mentions that est. cost doesn't include those.

FEATAnthropic models publish a one-time prompt cache write price, separate from the per-token cache-read rate, and it differs by TTL. Turn on Show cache prices and the violet tag shows both the 5-minute and 1-hour figures.

FEATSome models have a real access gate beyond per-token pricing, like Z.ai's glm-5 requiring an active GLM Coding Plan subscription. The fuchsia tag surfaces that note.

FEATAdded support for tiered, whole-request pricing. Some vendors (xAI and Google) reprice the entire request once input tokens cross a threshold (>200k). Cost estimates now account for this.

FEATThe Max output column now shows a Δ (delta) marker when a model's real ceiling is higher than the API's min. default output length.

FIXRenamed the "Cache-hit prices" toggle to "Show cache prices" for clarity.