Pricing

ALLM turns provider-specific rate cards into deterministic quotes without floating-point rounding.

Units and precision

Rate cards store prices as integer micro-USD per one million tokens. Quotes return exact decimal strings with six fractional digits. Parse total_usd as a decimal value or retain it as a string—never convert money to binary floating point for billing.

input_micro_usd
integerrequired
Micro-USD charged per one million non-cached input tokens.
cached_input_micro_usd
integer
Micro-USD charged per one million cached input tokens; falls back to the input rate when absent.
output_micro_usd
integerrequired
Micro-USD charged per one million output tokens.

Quote formula

Formula
regular_input = input_tokens - cached_input_tokens

total =
  regular_input × input_rate / 1,000,000
  + cached_input_tokens × cached_input_rate / 1,000,000
  + output_tokens × output_rate / 1,000,000

Cached input

cached_input_tokens is a subset of input_tokens, not an additional amount. For GPT-5.1, 1,000,000 input tokens of which 800,000 are cached produces 0.350000 USD before output: 200,000 at $1.25/M plus 800,000 at $0.125/M.

Missing pricing

Some image, audio, search, or newly released deployments use dimensions that the token quote endpoint does not yet normalize. Filter with has_pricing=true for token-priced deployments. A request without an applicable rate card returns 422 PRICING_UNAVAILABLE.