What does tool calling add to my API bill?

Updated October 2026 · How we answer

Short answerTool definitions and returned results both count as input or output tokens, so tool calling usually raises costs. The extra amount depends on how many tools you send and how large their outputs are.

Where the cost comes from

Each request that includes tool schemas sends extra text to the model, which adds input tokens. A long list of tools with detailed descriptions can add noticeable cost to every call. Results you send back also count, especially when a tool returns a large document or table. Output tokens for the model's own tool requests add a little more.

Multi-step flows add cost too. Each round trip includes the conversation so far, so a chain of calls can grow faster than expected.

Keeping spending predictable

Send only the tools relevant to each request instead of your full catalog. Trim tool outputs to the fields the model needs. Set a maximum number of tool rounds so a runaway loop cannot keep billing.

Check your provider's current pricing page for token rates and any tool-related fees. Pricing changes often, so do not rely on old numbers. Estimate the cost of a typical flow before you launch the feature.

  • Send only relevant tools for each request.
  • Return trimmed, structured results.
  • Cap the number of tool rounds per request.
  • Track token use per feature in your logs.

Common mistakes

  • Sending the entire tool catalog with every request.
  • Returning full raw API responses when the model only needs a few fields.
From our shopsCASEONYX: Dark-luxe tough phone cases.