Exporting Billing Metrics

Overview

Fireworks provides a CLI tool to export comprehensive billing metrics for all usage types including serverless inference, on-demand deployments, and fine-tuning jobs. The exported data can be used for cost analysis, internal billing, and usage tracking.

This export reports metered quantities — tokens and accelerator-seconds — not dollars. For a CSV of rated serverless costs grouped by model, model tier, user, or API key, use Exporting Usage Costs.

Use the Fireworks CLI to export a billing CSV that includes all usage:

# Authenticate (once)
firectl login

# Export billing metrics to CSV
firectl billing export-metrics

Examples

Export all billing metrics for an account:

firectl billing export-metrics

Export metrics for a specific date range and filename:

firectl billing export-metrics \
  --start-time "2025-01-01" \
  --end-time "2025-01-31" \
  --filename january_metrics.csv

Output format

The exported CSV includes the following columns:

email: Account email
start_time: Request start timestamp
end_time: Request end timestamp
usage_type: Type of usage (e.g., TEXT_COMPLETION_INFERENCE_USAGE)
accelerator_type: GPU/hardware type used
accelerator_seconds: Compute time in seconds
base_model_name: The model used
model_bucket: Model category
parameter_count: Model size
prompt_tokens: Input tokens
completion_tokens: Output tokens
cached_prompt_tokens: Prompt tokens served from cache (text inference only). Subset of prompt_tokens.
uncached_prompt_tokens: Prompt tokens not served from cache (text inference only). prompt_tokens - cached_prompt_tokens.

Older usage records and non-text usage types may not have a cached/uncached split in the underlying data. Exports normalize these rows to cached_prompt_tokens=0 and uncached_prompt_tokens=prompt_tokens, so prompt_tokens = cached_prompt_tokens + uncached_prompt_tokens always holds.

Sample row

email,start_time,end_time,usage_type,accelerator_type,accelerator_seconds,base_model_name,model_bucket,parameter_count,prompt_tokens,completion_tokens,cached_prompt_tokens,uncached_prompt_tokens
user@example.com,2025-10-20 17:16:48 UTC,2025-10-20 17:16:48 UTC,TEXT_COMPLETION_INFERENCE_USAGE,,,accounts/fireworks/models/llama4-maverick-instruct-basic,Llama 4 Maverick Basic,401583781376,803,109,200,603

Automation

Each firectl billing export-metrics call supports a maximum 31-day time range. To export longer historical ranges, run the command in multiple 31-day chunks and combine the CSV files in your downstream pipeline. You can automate exports in cron jobs and load the CSV into your internal systems:

# Example: Daily export with dated filename
firectl billing export-metrics \
  --start-time "$(date -v-1d '+%Y-%m-%d')" \
  --end-time "$(date '+%Y-%m-%d')" \
  --filename "billing_$(date '+%Y%m%d').csv"

# Example: Backfill 6 months in 31-day chunks
start_date="2025-01-01"
end_date="2025-07-01"
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-metrics \
    --start-time "$current_start" \
    --end-time "$current_end" \
    --filename "billing_${current_start}_to_${current_end}.csv"

  current_start="$current_end"
done

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

Coverage

This export includes:

Serverless inference: All serverless API usage
On-demand deployments: Deployment usage (see also Exporting deployment metrics for real-time Prometheus metrics)
Fine-tuning jobs: Fine-tuning compute usage
Other services: All billable Fireworks services

For real-time monitoring of on-demand deployment performance metrics (latency, throughput, etc.), use the Prometheus metrics endpoint instead.

Get Started

Serverless Inference

Deployments

Models & Inference

Training

Managed Fine-Tuning

Training API

Fire Pass

Fireworks Nexus

Administration

Security & Compliance

Integrations

Reference

Exporting Billing Metrics

Overview