Struct ModelPricing 

pub struct ModelPricing {
    pub input_cost_per_token: f64,
    pub output_cost_per_token: f64,
    pub cache_read_input_token_cost: f64,
    pub cache_creation_input_token_cost: f64,
    pub cache_creation_input_token_cost_above_1hr: f64,
    pub output_cost_per_reasoning_token: f64,
    pub web_search_cost_per_query: f64,
    pub tiers: Vec<ThresholdTier>,
    pub ranges: Option<Vec<TierRange>>,
}

Pricing data for a single AI model in USD per token.

Supports three strategies, checked in this order by calculate_cost:

1. Range-based (ranges is Some): input_tokens selects a TierRange and its prices are applied standalone. Used by Qwen / doubao families.
2. Threshold-based (tiers is non-empty): the highest ThresholdTier whose threshold_tokens is exceeded by total input context wins; all token types switch to that tier’s prices. Used by Claude Sonnet 4.x, Gemini 2.5 Pro, Gemini 1.5 (128k), GPT-5.x (272k), etc.
3. Flat (neither set): base prices apply to every request.

This struct is only ever held in memory — tiers / ranges are derived from the raw *_above_Nk_tokens / tiered_pricing keys of LiteLLM by parse_litellm_entry. Cache files store the raw LiteLLM cost fields verbatim (see filter_cost_fields); reloading the cache runs them back through parse_litellm_entry so the derived structures are reconstructed freshly on every launch.

Fields

input_cost_per_token: f64

Base input price in USD per token.

output_cost_per_token: f64

Base output price in USD per token.

cache_read_input_token_cost: f64

Base cache-read price in USD per token.

cache_creation_input_token_cost: f64

Base cache-write (5-minute TTL) price in USD per token.

cache_creation_input_token_cost_above_1hr: f64

Price per token for cache writes using Anthropic’s extended (1 hour) TTL. 0.0 means the model doesn’t support 1hr cached writes — callers fall back to cache_creation_input_token_cost (5-minute TTL price).

output_cost_per_reasoning_token: f64

Price for reasoning / thinking tokens emitted as part of the assistant response but billed separately from regular output tokens. Populated by Gemini 2.5 flash / flash-lite (thoughts_tokens), perplexity sonar-deep-research, and some qwen-turbo entries. 0.0 means the model doesn’t split reasoning from output — callers fall back to output_cost_per_token.

web_search_cost_per_query: f64

Price in USD for one server-side web-search request (server_tool_use.web_search_requests), billed per query rather than per token. Derived by parse_litellm_entry from LiteLLM’s nested search_context_cost_per_query object (Anthropic charges a flat $0.01 across its low/medium/high sizes). 0.0 means the model publishes no web-search price.

tiers: Vec<ThresholdTier>

Threshold-based tiers, sorted ascending by threshold_tokens.

ranges: Option<Vec<TierRange>>

Range-based pricing (mutually exclusive with tiers in practice).

Trait Implementations

impl Clone for ModelPricing

fn clone(&self) -> ModelPricing

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for ModelPricing

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for ModelPricing

fn default() -> ModelPricing

Returns the “default value” for a type.

impl<'de> Deserialize<'de> for ModelPricing

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for ModelPricing

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.