Crate openai_cost 

openai-cost

Calculate OpenAI API call cost from a usage block.

OpenAI returns three token counts on a chat-completion or responses-API response: prompt_tokens (a.k.a. input_tokens), completion_tokens (a.k.a. output_tokens), and prompt_tokens_details.cached_tokens (a.k.a. cached_input_tokens on the Responses API). Each has its own price. This crate gives you a small Pricing table for popular models and a Usage struct that knows how to compute cost from those fields.

Pricing is best-effort and dated; verify against https://openai.com/api/pricing/ before using these numbers for billing.

Quick example

use openai_cost::{Usage, default_pricing};

let pricing = default_pricing("gpt-5").unwrap();
let usage = Usage {
    input_tokens: 1_000,
    output_tokens: 500,
    cached_input_tokens: 0,
};
let cost = pricing.cost_for(&usage);
// 1k * input + 500 * output, divided by 1M
assert!(cost > 0.0);

Versioned model ids

use openai_cost::default_pricing;
// The crate normalizes dated suffixes (`gpt-5-2026-04-01`) back to
// the base model name.
assert!(default_pricing("gpt-5-2026-04-01").is_some());

BYO pricing

use openai_cost::{Pricing, Usage};
let custom = Pricing {
    input_per_mtok: 1.25,
    output_per_mtok: 5.0,
    cached_input_per_mtok: 0.125,
};
let usage = Usage::default();
let _ = custom.cost_for(&usage);

Structs

Pricing

Per-model rates, USD per 1M tokens.

Usage

Three-field token usage as returned by OpenAI’s chat-completion and responses APIs.

Constants

DEFAULT_PRICING_TABLE

Built-in pricing table. Source: openai.com/api/pricing as of 2026-Q2. VERIFY before billing.

Functions

default_pricing

Look up the price table entry for an OpenAI model id.

normalize_model_id

Strip a trailing -YYYY-MM-DD snapshot suffix.