1. Pricing sources
Model prices are entered from first-party provider documentation only. Every model record stores a pricing source, verification date, effective date, lifecycle status and service-tier rates.
Our goal is not false precision. Every estimate separates verified price data, user assumptions and uncertain tokenization.
Model prices are entered from first-party provider documentation only. Every model record stores a pricing source, verification date, effective date, lifecycle status and service-tier rates.
Time-limited prices are represented as dated schedules. For example, an introductory rate can automatically change on its published effective date rather than silently becoming stale.
Text pasted into the site is estimated using word, UTF-8 byte, script and code-density signals. This is explicitly labeled as an estimate. Exact provider-reported tokens always override it.
When a provider documents a token threshold, the engine applies the higher band to the full request once the threshold is crossed. The applied band appears in the result.
Uncached input, cache hits, cache writes and storage are calculated separately. Unsupported cache dimensions fall back to standard input only when the provider has no separate published rate.
Published per-call fees are modeled separately from token costs. Agent calculations expose context growth by step instead of hiding it inside a single multiplier.
total = uncached input + cached input + cache writes + output + cache storage + regional uplift + tool fees
All token prices are normalized to USD per one million tokens before multiplication.
When a verified error is reported, we correct the registry, add an entry to the pricing changelog and retain the effective date. Contact with the model ID, official source and disputed value.