InferCost · docs

Cloud pricing catalog

InferCost ships with a verified catalog of list prices for OpenAI, Anthropic, and Google models so cloud comparisons work out of the box. The catalog is a single YAML file, overridable per-cluster when you have negotiated enterprise rates.

The catalog file

Canonical location: config/pricing/cloud-pricing.yaml. A byte-identical copy is embedded into the controller binary at build time so the Go code and the YAML can never silently disagree.

lastVerified: 2026-03-21

sources:
  OpenAI: https://developers.openai.com/api/docs/pricing
  Anthropic: https://platform.claude.com/docs/en/about-claude/pricing
  Google: https://ai.google.dev/gemini-api/docs/pricing

providers:
  - provider: OpenAI
    models:
      - model: gpt-5.4
        tier: flagship
        inputPerMillion: 2.50
        outputPerMillion: 15.00
      # ...

What list prices do and do not reflect

They are the published public rates on the provider's pricing page. They do not reflect:

  • Negotiated enterprise / committed-use agreements
  • Batch API discounts (typically 50% off)
  • Prompt caching discounts
  • Azure OpenAI / Bedrock / Vertex markup or rebate
  • Free-tier allocations

For any team with a meaningful cloud contract, the comparison in InferCost is therefore an upper bound on what you would pay going cloud. Override the catalog with your real rates to make the comparison honest both ways.

Overriding with your negotiated rates

In Helm values:

pricing:
  override:
    enabled: true
    inline: |
      lastVerified: 2026-07-01
      sources:
        OpenAI: "Negotiated agreement with OpenAI 2026-Q2"
      providers:
        - provider: OpenAI
          models:
            - model: gpt-5.4
              inputPerMillion: 1.75      # negotiated 30% off list
              outputPerMillion: 10.50

The chart mounts the override as a ConfigMap at /etc/infercost/pricing/cloud-pricing.yaml and passes --pricing-file to the controller. The controller logs which catalog is in use at startup:

INFO  loaded pricing override   lastVerified=2026-07-01 providers=1 path=/etc/infercost/pricing/cloud-pricing.yaml

A malformed or missing override file causes the controller to fail at startup rather than silently fall back to list prices — if you've configured custom rates, we owe you a loud error when we can't use them.

Refreshing the catalog

The lastVerified field is a contract. When it's older than ~3 months, the catalog should be refreshed even if nothing visibly changed — "confirmed no change since X" is still useful. The procedure is one PR, documented in docs/pricing-refresh.md. Pull requests to refresh the default catalog are welcome.

Schema validation

The loader runs Validate() on every file it reads. The three failure modes it catches, because each of them silently produces wrong cost comparisons:

  • Empty or missing providers
  • Duplicate (provider, model) pairs
  • Negative pricing values

If validation fails on an operator override, the controller logs the specific problem and refuses to start. If validation fails on the embedded catalog, CI catches it before the binary ships.