bedrock-cost

AWS Bedrock invocation cost across vendors (Llama, Mistral, Cohere, Titan, AI21). Cross-region inference profile aware (us., eu., apac.). Zero deps.

For Anthropic Claude models on Bedrock, use claude-cost — it handles the extra cache_creation / cache_read fields.

Usage

use bedrock_cost::{Usage, default_pricing};

let p = default_pricing("meta.llama3-70b-instruct-v1:0").unwrap();
let u = Usage { input_tokens: 1_000_000, output_tokens: 500_000 };
let cost_usd = p.cost_for(&u);

Model id normalization

use bedrock_cost::default_pricing;
// Cross-region inference profile
assert_eq!(
    default_pricing("us.meta.llama3-70b-instruct-v1:0"),
    default_pricing("meta.llama3-70b-instruct-v1:0")
);
// Full ARN
assert!(default_pricing("arn:aws:bedrock:us-east-1::foundation-model/cohere.command-r-plus-v1:0").is_some());

Pricing notes

All rates are USD per 1,000,000 tokens (us-east-1 region) as of 2026-Q2. Bedrock prices vary by region — verify against https://aws.amazon.com/bedrock/pricing/ before billing.

License

MIT or Apache-2.0.