AI Bill and billing imports

Where billing data comes from

Provider-reported billing is stored separately from usage metered by the SDK, the router or OpenTelemetry, and is never merged with it. The AI Bill picks one source per provider, as described in How the total is worked out.

Fireworks costs are rated costs from its usage reports, not a settled invoice. They count toward the total like other account reports.

Connect a provider account

1. Open Settings › Providers and choose the Connect account tab (or Connect an account while nothing is connected).
2. Pick the Provider. The form shows that provider's reporting tier and what it cannot report.
3. Enter a Connection name, the provider's account identifier and the credential (see the table below). Connect each provider account once.
4. Choose Connect and sync. Token Meter checks the credential with the provider before saving, stores it encrypted, and queues the first import.

The cloud usage permissions above are optional. Without them the billing sync still works and the connection says token usage is unavailable.

Manage connections

The Connections tab of Settings › Providers shows one card per connection, with its status and the history imported so far.

• View history opens that connection in Data health › Provider usage.
• Sync now queues a sync. If one is already queued or running, nothing new is queued.
• Edit credentials (in the ⋯ menu) rotates the key. The provider cannot be changed; enter the credential again and choose Save credentials and sync. Rotating never counts against a plan's connection cap.
• Backfill history re-imports the provider's full available history. Rows already imported are updated in place, not duplicated. It uses provider API quota.
• Disconnect asks you to type the connection name. The stored credential is erased and imported history is kept. Reconnect opens the edit form, where you enter the credential again.
• Backfills and disconnects are recorded in the workspace's audit log.

Manual bill import (CSV)

Use it for a provider without an automatic connection, or to load an invoice you already have. It is on Settings › Providers, below the connections.

billing_day,sku,amount_usd,model_id,token_class,quantity
2026-10-01,Input tokens,12.40,llama-3.3-70b,input,15500000
2026-10-01,Output tokens,8.10,llama-3.3-70b,output,2700000

1. Choose the Provider and the file (at most 5,000 rows and 2 MB), then Upload & preview.
2. The preview shows total, valid and invalid rows, the total amount of the valid rows, each invalid row with its reason, and a sample of the rows to import. Nothing is in your AI Bill yet.
3. Choose Confirm import to write the valid rows, or Discard. Invalid rows are skipped; fix them in the file and upload again.

• Recent uploads are listed with their state: Awaiting confirm, Processing, Imported or Failed.
• Uploading the exact same file again opens the earlier upload instead of creating a new one.
• A row identical to one already imported (same provider, day, SKU, model, token class, project or key and amount) is updated, not added twice.
• Imported rows count as that provider's invoice in the AI Bill. They are not reconciled against metered spend.

How the total is worked out

Total AI spend, this range at the top of Analytics › Bill covers the chosen range (This month, Previous month, 30d or 90d, in UTC days). Each line is shown separately:

So each provider counts once: its invoice, else its account report, else metered. A provider is never counted from two sources for the same day.

• An invoice covers the whole range. When a provider has any invoice line in the range, its invoice total is used and metered spend is not added, even for recent days the invoice has not reached yet.
• Account reports count whole periods. Only report periods that lie entirely inside the range are added; a period is never prorated.
• Coding tools. Claude Code and Claude chat on API keys are billed by Anthropic, Gemini CLI by Google, and Codex on an OpenAI API key by OpenAI. When that provider's invoice is imported for the range, the tool's spend is already inside it, so it is left out and shown as Coding-tool spend already inside a provider invoice (disclosure). Codex from the ChatGPT Enterprise feed, and tools billed by no imported provider, are always added.

What the total leaves out

The page lists these under the total:

• Cursor usage-based spend, which is not collected yet.
• Expense imports. They are used to find unmanaged AI tools, and they repeat bills already counted here.
• The API-equivalent value of subscription use, which is not money spent.
• Account reports with no US dollar amount (Mistral, DeepInfra, or a report in another currency). The provider is named in the note.

Google coverage is partial. With a Google connection, per-model usage comes from Vertex AI Cloud Monitoring only. Gemini Developer API (AI Studio) traffic is not in metered estimates or reconciliation unless you send it through the SDK or an observed gateway; the Cloud Billing export may still include its spend.

The provider table

Below the total, three tiles show Metered spend, Provider invoice lines (with the number of imported lines) and Drift. The table has one row per provider; the provider name opens its requests. It shows every source side by side and never adds them up.

Unmapped spend is expected for some line items: Google cache storage, audio, image, video, tuned models and SKUs shared by two models; Bedrock models other than Claude; Azure fine-tuning, realtime, audio, image and preview meters.

Reconciliation

Every day at 07:00 UTC, Token Meter compares metered spend with invoice lines from connected importers, per provider and day. Drift is the invoice minus metered spend. A day is within tolerance when the drift is at most 2% of the invoice; a day with no invoiced amount matches only if metered spend is also zero.

• The Drift column reads no invoice import without invoice lines, not reconciled when lines exist but no day has been compared (CSV uploads are never reconciled), - under 50 cents, and is amber when outside tolerance.
• To make metered spend match a negotiated contract, add a private rate under Settings › Providers. It outranks the public list price for this workspace; a start date up to 365 days back reprices usage from that day.
• Azure OpenAI subscriptions billed in a currency other than US dollars are not reconciled.

Contracts and credits

The Contracts tab of Analytics › Bill tracks commitments, credits and settlement state separately from metered usage and invoices. Nothing there rewrites an imported invoice line.

• Add a contract tracks a minimum-spend commitment against your imported invoice totals for the contract's window. Record a period's consumption (blank amount imports the invoice lines) and any separately billed overage. Overage enters the total only when Overage already appears in provider invoice is unticked.
• Add a credit or adjustment: a Credit reduces spend; an Adjustment goes either direction. It applies on the date you give and is subtracted once from the AI Bill total.
• Set settlement state records accounts-payable status (invoiced, paid, disputed) for one provider and month.

Data health › Provider usage

Everything the connections imported, up to 90 days, shown separately from metered estimates and from invoice reconciliation. Missing measurements read Unavailable; nothing is inferred.

Evidence labels: SDK-metered is captured from an instrumented request, Provider-reported comes from the provider's usage or cost API and may be rated or provisional, and Invoice-confirmed matches a settled invoice or verified invoice import.

For the per-call metering behind the metered estimate, see the Quickstart and Providers.

Troubleshooting