Start here

Choose a guide based on what you need to do. You can look up prices without integrating an SDK, connect usage tracking to an existing app, or follow a process for reducing production costs. The sidebar groups these tasks into the paths below.

​

Estimate costs

Use these guides when you need to know what a request could cost or compare model and provider rates. Start with searching model prices if you need the correct identifiers. If you already know your model, provider, and token counts, go directly to calculating a request’s cost. This path uses the Narev Cloud Pricing API. It includes provider comparisons and a DeepSeek example that explains cached-token costs. You don’t need to run inference or install the Narev SDK.

​

Track usage and bill customers

Choose the integration that matches your existing stack:

• DeepSeek usage tracking: record usage and cost as your app generates responses.
• Langfuse cost backfilling: add cost estimates to generations you already recorded.
• Chatbot with Polar billing: connect customer identity, usage events, and a billing meter.

Start by verifying the recorded cost. Before charging customers, also verify the customer mapping and meter units.

​

Reduce AI costs

Start with the FinOps for AI workflow when you need a plan for an existing workload. It takes you through setting objectives and ownership, understanding spending, and testing changes before releasing them. The model-switch and prompt-engineering studies in this section show how to interpret cost and quality tradeoffs. Use them when you’re ready to evaluate a specific change.

​

Billing concepts

Read AI pricing models when you’re deciding what customers should pay for. Continue with usage-based billing to understand how measured usage becomes a customer charge.