Exporting Usage Costs

Overview

Fireworks provides a CLI tool to export rated dollar costs for serverless inference as a CSV. Rows are bucketed by UTC day and grouped by one dimension you choose — model, model tier, user, or API key — which makes it the fastest way to produce a chargeback or per-team spend report. The export is the CSV counterpart of POST /usageCosts:query.

This page — rated dollar costs for serverless inference, as a CSV grouped by model, model tier, user, or API key.
Exporting Billing Metrics — metered quantities (tokens, accelerator-seconds) for every usage type, one CSV row per usage event. No dollars.
Usage & Cost Breakdown — the same data as ad-hoc CLI or HTTP API queries rather than a CSV, with richer grouping and filtering.

This export reports costs for the whole account, so it requires account administrator access. Non-admin users can read their own costs through POST /usageCosts:query at SELF scope.

# Authenticate (once)
firectl login

# Export usage costs to CSV
firectl billing export-usage-costs

With no flags, this exports the last 24 hours grouped by model and writes usage_costs.csv to the current directory.

Examples

Group costs by API key instead of model:

firectl billing export-usage-costs --group-by api_key

Export a specific date range, dimension, and filename:

firectl billing export-usage-costs \
  --start-time "2026-05-01" \
  --end-time "2026-06-01" \
  --group-by user \
  --filename may_costs_by_user.csv

Output format

The exported CSV always has these six columns, regardless of which dimension you group by:

account_id: Your Fireworks account ID
date: UTC day bucket, as YYYY-MM-DD
group_by_dimension: The dimension the row is grouped by — model, model_tier, user, or api_key
group_by_value: The value for that dimension (see below)
subtotal_usd: Rated cost for that day and value, as an exact decimal with nine fractional digits
currency: Currency code for subtotal_usd, for example USD

group_by_value depends on --group-by:

`--group-by`	`group_by_value`
`model` (default)	Model resource name, e.g. `accounts/my-account/models/glm-5p2`
`model_tier`	Serving capability tier used to price the usage, e.g. `GLM 5.2 (Fast)`
`user`	User resource name, e.g. `accounts/my-account/users/alice`
`api_key`	Stable API key ID, e.g. `key_4nMFyHCSZP4CRKqa`. Never plaintext key material

Two reserved values can appear in group_by_value: unattributed, when the underlying billing event carries no value for the requested dimension, and unknown_model, when a billing model can’t be mapped to a public model. Days with an exact-zero subtotal are omitted rather than written as 0.000000000 rows, so a dimension value only appears on days it actually cost money.

Sample row

account_id,date,group_by_dimension,group_by_value,subtotal_usd,currency
my-account,2026-05-14,model,accounts/my-account/models/glm-5p2,8.400000000,USD

Automation

Each firectl billing export-usage-costs call supports a maximum 31-day time range, and --start-time cannot be more than 100 days in the past. To export longer historical ranges, run the command in multiple 31-day chunks and combine the CSV files in your downstream pipeline.

# Example: Daily export with dated filename
firectl billing export-usage-costs \
  --start-time "$(date -v-1d '+%Y-%m-%d')" \
  --end-time "$(date '+%Y-%m-%d')" \
  --group-by api_key \
  --filename "usage_costs_$(date '+%Y%m%d').csv"

# Example: Backfill 90 days in 31-day chunks
start_date="2026-03-01"
end_date="2026-06-01"
current_start="$start_date"

while [ "$(date -j -f "%Y-%m-%d" "$current_start" "+%s")" -lt "$(date -j -f "%Y-%m-%d" "$end_date" "+%s")" ]; do
  current_end="$(date -j -v+31d -f "%Y-%m-%d" "$current_start" "+%Y-%m-%d")"

  # Clamp the chunk end to the requested end_date
  if [ "$(date -j -f "%Y-%m-%d" "$current_end" "+%s")" -gt "$(date -j -f "%Y-%m-%d" "$end_date" "+%s")" ]; then
    current_end="$end_date"
  fi

  firectl billing export-usage-costs \
    --start-time "$current_start" \
    --end-time "$current_end" \
    --filename "usage_costs_${current_start}_to_${current_end}.csv"

  current_start="$current_end"
done

Run firectl billing export-usage-costs --help to see all available flags and options.

Coverage

Serverless token usage only. Costs are priced from cached input, uncached input, and output tokens. Dedicated deployment and fine-tuning spend are not included — for those, use Exporting Billing Metrics or the account-level totals from firectl billing get-usage --account-costs-only.
Subtotals are rated, not invoiced. They price usage with the subscription prices that apply to your account and exclude fixed fees, invoice-level discounts, minimums, credits, and taxes, so they may differ from the final invoice.
One dimension at a time. Rows are always bucketed by day plus the single dimension passed to --group-by. To combine dimensions, use POST /usageCosts:query, which accepts up to two.

Get Started

Serverless Inference

Deployments

Models & Inference

Training

Managed Fine-Tuning

Training API

Fire Pass

Fireworks Nexus

Administration

Security & Compliance

Integrations

Reference

Exporting Usage Costs

Overview