Pricing methodology

The formula

For token-metered usage, each token type is priced independently against a per-million rate, then summed into an API-equivalent cost:

inputCost      = inputTokens      / 1e6 × inputRate
outputCost     = outputTokens     / 1e6 × outputRate
cachedInputCost= cachedInputTokens/ 1e6 × cachedInputRate
cacheWriteCost = cacheWriteTokens / 1e6 × cacheWriteRate
reasoningCost  = reasoningTokens  / 1e6 × reasoningRate

apiEquivalent  = inputCost + outputCost + cachedInputCost
               + cacheWriteCost + reasoningCost

Decimal-safe math

All money arithmetic runs through arbitrary-precision decimals, never floating point. A sub-cent per-token cost is preserved exactly, so a real $0.0004 is never rounded away to $0.00.

Historical pricing

Pricing records are versioned with an effective date range. A project built in the past is priced with the record that was in effect on its build date. If we don’t have a record for that date, we say so — we never silently fall back to today’s rate.

When we show a range

Precise token math with a known model gives a single figure. We widen to a range when:

• Only a lump total-token count is available (no input/output split).
• The cost comes from a subscription allocation rather than metered usage.
• Credits must be converted through an assumed credit value.

Seed pricing disclaimer

The pricing records shipped for demonstration are published-rate approximations, each carrying a note, and must be verified in the admin pricing tool before they drive any real reporting. TokenSpent never presents unverified pricing as authoritative.