Skip to main content
Baseten runs serverless model APIs priced per token, dedicated model serving priced per minute of deployment uptime, and training. Input, cached input, and output each carry their own rate. CostGraph reads Baseten as FOCUS 1.2 records, one per priced unit: per model per token type per day for model APIs, and per deployment per day for dedicated serving.

​
Choose how the data reaches us

CostGraph pulls it

You supply a read-only key and we call Baseten once a day. Nothing to run.

You push it

You send us FOCUS records over the API. Credentials never leave your infrastructure.

​
Before you begin

  • In the Baseten console, go to Settings > API keys and create a key. It reads your whole organization’s billing, not just your own usage.

​
Let CostGraph pull

  1. Open Settings > Integrations and pick Baseten.
  2. Name the connection and pick the tenant these charges belong to.
  3. Enter the key you copied.
  4. Select Connect.
CostGraph verifies the credentials with a short fetch before saving, so a wrong value or a missing permission fails while you are still on the form.
Credentials are encrypted at rest and used only to read billing data. Revoke them in Baseten at any time and the connection stops; nothing else is affected.

​
Push it yourself

Send the charges to CostGraph as FOCUS records instead. Push FOCUS data covers the connection id, the focus:write key, and the request shape.

​
Run the exporter

Rather than build that request yourself, run our exporter. It reads Baseten on a schedule and pushes the records for you, wherever you want to run it: a container, a Kubernetes CronJob, or a binary on a machine you already have.
Run the exporter covers the three ways to run it and the variables it reads.

​
Baseten behavior

Baseten’s billing summary reports token counts and rounds its own money to whole cents, which reads as zero for anything smaller. CostGraph prices each token type from the rates Baseten publishes for that model instead, so a fraction of a cent still lands as a real figure. Cached input is priced well under uncached input. A model that is not in Baseten’s catalogue is skipped rather than priced at a guessed rate, and billing lags behind usage, so a sync run immediately after traffic can read empty.

​
What CostGraph does with it

Records land as raw billing rows, are normalised into line items, and roll up into daily cost. From there Baseten spend appears in Cost Overview and in anomaly detection alongside every other provider. Two signals ride along:
  • Cache savings. Cached input is much cheaper than uncached, and Baseten reports the split, so you can see how much of a bill the prompt cache avoided.
  • Idle deployments. A dedicated deployment reports billed minutes and the requests it served on the same row, so one that is paid for and barely used stands out.