01
Customer API-credit usage
Total input tokens are multiplied by the published customer input rate; output tokens are multiplied by the published output rate. The sum reduces the credit balance.
Pricing methodology
How ScaleMax separates customer billing, token accounting, and internal provider cost.
Every completed request records an immutable usage snapshot. Customer charges use published ScaleMax rates, token keys use raw provider-reported token totals, and internal provider cost remains a separate administrative ledger.
Three separate records
One request, three different questions.
02
Token-budget usage
Provider-reported total input, including cache buckets where reported, is added to output once. That provider-total token count reduces the prepaid allowance.
03
Internal provider cost
ScaleMax separately records configured provider input, cache, cache-write, output, and total cost. These internal rates are not used as the customer API-credit charge.
API-credit formula
Published input plus published output.
Customer billing does not apply an internal provider cache discount. Cached and cache-write tokens remain part of total customer input when the provider reports them.
frontend cost =
(total input ÷ 1,000,000 × published input rate)
+
(output ÷ 1,000,000 × published output rate)Example using GPT-5.6 Terra: 1M input × $2 plus 0.2M output × $12 = $4.40 in usage credits.
With the Starter pack’s 10× credit value, that is approximately $0.44 of the pack’s base purchase amount when the balance is fully used.
Token-budget formula
Provider-total input plus output.
ScaleMax preserves cache buckets without double-counting them. If the provider supplies a total input count, that total is the floor; if it supplies separate uncached, cache-read, and cache-write buckets, the normalized total includes all reported buckets.
provider total tokens =
normalized total input tokens
+ output tokensExample: 40M uncached input + 60M cached input + 20M output consumes 120M token-budget tokens. A cache bucket is never silently dropped.
Stored at request time
Historical usage does not change when a future price changes.
Effective-dated rates
The pricing engine loads the rate snapshot effective for the request. A later price update applies to later requests.
Integer money units
Charges are calculated in integer microdollars to avoid floating-point drift, then formatted as decimals for display.
Large-context tiers
When a model has a configured context threshold and long-context rates, the pricing engine selects that tier automatically.
Unavailable cache metadata
If a provider does not report cache counts, ScaleMax marks cache status unavailable rather than inventing a number.
Use the live catalog for current rates.
This page explains the calculation. The live public model catalog is the source for currently routed model IDs and customer-facing prices.