Costs
How the per-minute estimate is worked out, how each call's real cost is recorded, and where to see both.
What you will achieve
You will know what an agent is expected to cost, and where to see what a call actually cost.
The estimate
On an agent's page, the Cost card shows "$X per minute" for its current model, voice and hearing choices, with a line such as "$0.356 spent over 7 calls". The estimate assumes a typical minute of conversation: about 590 characters spoken by the agent, about 3,800 tokens read by the model and about 150 produced.
Real calls vary. A longer prompt, more knowledge lookups or a chatty caller moves the figure. The panel on the right of each choice shows the working.
The public Pricing page has a calculator that lets you compare models, voices and hearing options and see monthly totals.
What each call actually cost
Open History, then a call. The call page shows its cost split into the parts that made it up:
- the language model (tokens in, cached tokens and tokens out),
- hearing (seconds of speech recognised),
- voice (characters spoken),
- knowledge lookups (tokens embedded),
- the post-call review (the summary and outcome), and
- the platform's per-minute charge.
The cost is stored with the call using the prices at the time, so past calls do not change when prices do.
Where totals appear
- Agents page: Total spent across all agents.
- Analytics: cost per call and total cost over a date range.
- Knowledge: the Total spent column shows what each base's embeddings cost.
Configuration details
Choosing a cheaper model, a faster voice or a shorter prompt lowers cost. Test calls are billed like any other call.
Related guides
Something here does not match what you see in the dashboard? Let us know.