Transparent AI costs
See what each AI call cost, down to the request
Every Claude call GetFinOps makes for your workspace is stored with its tokens, cache use, list-price cost, stop reason and Anthropic request ID, and listed call by call.
It does not show Anthropic’s invoice.
The problem
When a product uses AI on your behalf, its cost is easy to hide in one number. You can’t tell which feature spent it, whether prompt caching was priced, or which request a charge came from.
How it works
Record
When a Claude call returns, its usage is read from Anthropic’s response and stored with its cost, feature and request ID.
Inspect
The AI Spend page shows spend over time, by agent and by model, and each model opens the list of calls behind its total.
Total
Settings → Billing shows the month-to-date total of those calls, without Bedrock or Vertex AI spend, next to the included amount.
Request ID
from Anthropic, stored with each recorded call, plus why the response stopped
1.25 times
the model’s input rate for cache writes, and 0.1 times for cache reads
5 read tools
on the MCP server for LLM usage, down to the list of individual calls
Admin key
must be connected before figures are compared with Anthropic
Capabilities
Every Claude call made for your workspace, recorded with its tokens, cost and request ID.
A record for every call
Each Claude call made for your workspace stores the model, input, output and cache tokens, cost, duration, the feature that made it, the stop reason and Anthropic’s request ID, so the call can be traced on Anthropic’s side.
Cache tokens priced separately
Anthropic reports prompt-cache tokens apart from input tokens. GetFinOps prices cache writes at 1.25 times and cache reads at 0.1 times the model’s input rate, using Anthropic’s published list prices.
The calls behind a model’s total
On the AI Spend page, click a model in Cost by Model to list its calls, newest first, with the cost of each. Calls cut off at the token limit are flagged, and the totals cover every matching call.
Usage in Settings → Billing
Settings → Billing shows month-to-date AI cost, requests, the included amount and any overage. It counts the Claude calls GetFinOps makes for your workspace and leaves out Bedrock and Vertex AI spend collected from your clouds, which your cloud provider bills.
In depth
What is recorded for each call
Every Claude call GetFinOps makes for your workspace, including report narratives, the daily brief, the board builder and the chats, writes one usage row. The billing fields are read from the response Anthropic returns.
- The model that answered, and its input, output, cache-write and cache-read tokens.
- The cost in US dollars, to six decimal places.
- The feature and, where set, the agent that made the call, and how long it took.
- The stop reason, so a response cut off at the token limit is visible.
- Anthropic’s request ID, the key Anthropic can look a charge up by.
- For a per-account report narrative, the cloud account it was about. Workspace-wide work, such as chat, is stored without one.
How cost is calculated
Each model has input and output rates in dollars per million tokens, taken from Anthropic’s published pricing. Cache writes are priced at 1.25 times the input rate and cache reads at 0.1 times. Anthropic reports cache tokens separately from input tokens, so leaving them out would under-report cost.
A model missing from the rate card is priced at a default rate rather than dropped, so the call still counts. When the router chose the model, the row also stores what the same call would have cost on the default model, and the AI Spend page shows what routing saved or added.
Looking at individual calls
On the AI Spend page, click a model in Cost by Model to open its calls, newest first. Each row shows the time, feature, tokens including cache, cost, duration, stop reason and request ID. The header totals cover every call that matches the filter, and Load more fetches the next page.
Only calls GetFinOps sent are listed one by one. Bedrock and Vertex AI spend collected from your clouds arrives as daily totals, so it counts toward a model’s total but has no individual calls.
Usage in Settings → Billing
Settings → Billing has an AI usage card with month-to-date cost, request count, the included amount and any overage. It counts the Claude calls GetFinOps makes for your workspace; Bedrock and Vertex AI spend collected from your clouds is left out, because your cloud provider bills it. The hosted service doesn’t report this usage to Stripe yet.
Keeping models and prices current
The models the router may call are checked at startup against a committed catalogue of Anthropic model IDs, and any mismatch is logged. Retired models stay on the rate card for pricing only, so older rows keep their cost.
Once a day, a background job sends a one-token request to each catalogued model and reports any that Anthropic has retired. Those probe calls are recorded under a platform account, not a customer workspace.
Checking against Anthropic
Where an Anthropic organisation admin key is connected, a job every six hours takes the previous UTC day. It compares the recorded cost of every call GetFinOps sent that day, across all workspaces, with the cost Anthropic reports, and stores the difference as a percentage. By default, a difference above 2% is logged as a warning.
Bedrock and Vertex AI spend is left out of that comparison, because it is billed through AWS and Google Cloud. Without an admin key, each day is still stored with GetFinOps’ own figure, and Anthropic’s figure is marked unavailable.
What it does not do
- It does not show Anthropic’s invoice. Figures are list-price calculations from GetFinOps’ own records.
- It does not compare figures with Anthropic unless an organisation admin key is connected.
- It does not list individual Bedrock or Vertex AI calls, only their daily totals.
- The hosted service doesn’t report AI usage to Stripe yet.
- It does not send trial or free workspace usage to Stripe.
FAQ
Questions
Is this Anthropic’s invoice?
No. The figures are GetFinOps’ own records, priced at Anthropic’s published list rates. A model missing from the rate card is priced at a default rate rather than left out.
How are the figures checked against Anthropic?
Where an Anthropic organisation admin key is connected, a job compares each UTC day’s recorded cost with Anthropic’s cost report and stores the difference as a percentage. The comparison covers the whole Anthropic organisation, not one workspace. Without the key, no comparison is made.
Is AI usage reported to Stripe?
Not yet. The hosted service doesn’t send AI usage to Stripe today, for any plan. The Settings → Billing card still shows month-to-date usage, including a trial’s.
Can my own AI assistant read these numbers?
Yes. The MCP server has five read tools for LLM usage: a summary, totals by model and by agent, a time series, and the list of individual calls.
Try free — no credit card
Create a workspace and connect an AWS or GCP account, or book a demo first.