Module pricing 

Per-token price table keyed on each provider’s name()

Model Pricing

Compile-time pricing data for the providers this crate runs, keyed on the exact string each one’s LlmProvider::name reports. MetricsProvider prices every call through it; a consumer that layers operator overrides on top builds its own lookup from lookup_pricing, cost_from_pricing and zero_cost_for_unpriced, which are public for that purpose.

Cache accounting

A prompt token can bill at three rates, so the pricing entry carries the multipliers rather than a single global discount: fresh input at 1.0×, a cache read at ModelPricing::cache_read_multiplier (Anthropic 0.10×, OpenAI 0.50×, Gemini 0.25×), and a cache write at ModelPricing::cache_write_multiplier — which is a premium on Anthropic (1.25×), not a discount. Reasoning tokens bill at the output rate and are additive to the completion count, because every provider that reports them separately excludes them from it.

Counts arrive on TokenUsage, are carried through TokenCounts, and are priced by calculate_cost_for.

Structs

LongPromptRates

The rate card a model switches to when a request’s prompt is long.

ModelPricing

Per-model pricing rates in USD per million tokens, plus the two cache multipliers that turn a raw token count into a billed one.

TokenCounts

The token counts a single LLM call reports, in the shape they bill.

Constants

DEFAULT_CACHE_READ_RATE

Default fraction of the input rate at which a context-cache read bills.

DEFAULT_CACHE_WRITE_RATE

Default fraction of the input rate at which a context-cache write bills.

NOT_PER_TOKEN_METERED_PROVIDERS

Providers whose per-token cost is genuinely $0 by design, so a $0 billed cost is correct rather than a missing-price undercount.

PRICING_TABLE

Compile-time pricing table: (provider, model_prefix, pricing)

Functions

calculate_cost

Calculate the cost of an LLM request using compile-time pricing.

calculate_cost_for

Calculate the cost of an LLM request from a full TokenCounts.

calculate_cost_with_cache

Calculate the cost of an LLM request, breaking the prompt into cached + fresh tokens.

cost_from_pricing

Compute the USD cost from a resolved ModelPricing + token counts. Shared by the compile-time lookup and any override path a consumer layers on top, so both sources price a call identically.

is_not_per_token_metered

True when a provider bills via flat-rate subscription or self-hosting, so a $0 per-token cost is correct rather than a missing-price undercount.

lookup_pricing

Look up pricing for a (provider, model) pair using prefix matching.

pool_family

The provider a pooled tier prices as: claude-code#2 is a second Claude account, billed like the first. See crate::pool.

zero_cost_for_unpriced

Log the resolution of a (provider, model) pair that has no PRICING_TABLE or override entry, at a severity matching why it is unpriced, and return the $0 cost all three paths fall back to.