Loading...
Anthropic's family of large language models, priced per token and common in enterprise stacks.
Quick Definition
Claude is Anthropic's family of large language models, used widely in enterprises for coding, analysis, and content tasks. It is consumed through an API and through cloud platforms like Amazon Bedrock, and is priced per token, making token volume and model-tier choice the main cost levers.
Claude is Anthropic's family of large language models, widely used for coding assistance, document analysis, and customer-facing AI features. Teams consume it through Anthropic's API or through cloud platforms such as Amazon Bedrock, which makes Claude usage appear directly inside AWS bills.
Like other frontier models, Claude follows token-based LLM pricing: you pay per million input and output tokens, and rates differ sharply between model tiers. The smaller, faster tiers cost a fraction of the most capable ones, so matching the model tier to the task is the single biggest cost decision, many workloads run perfectly well on a smaller tier at a tenth of the price.
Example. A support automation team uses a top-tier Claude model for every ticket. An audit shows 80 percent of tickets are simple lookups; routing those to a smaller tier and reserving the large model for complex cases cuts the monthly model bill by 60 percent with no quality complaints.
From a FinOps view, Claude spend is managed like any usage-based service: track token consumption per feature, set budgets and alerts, and watch prompt length, since long prompts multiply input tokens. The Amazon Bedrock pricing guide explains the cloud-platform route, and the token budgeting guide covers controlling spend on any model.
Per token: input tokens (the prompt) and output tokens (the response) are metered separately, with rates depending on the model tier chosen.
Because it is available through Amazon Bedrock, where usage is billed by AWS like any other service, which also makes it visible to cloud cost tooling.
Use the smallest model tier each task allows, shorten prompts and system messages, and route only genuinely hard cases to the most capable tier.