About recoveryCompare recovery

Model Cost Breakdown

Validate model-level spend, request volume, and concentration flags from your trace data.

Results update live as you paste or type.

Live Preview

Paste model-level trace JSON above to see spend share, request volume, and concentration flags.

Why this tool

The pricing page says GPT-4 is $30/1M tokens, but your invoice says otherwise — because some tenants are on a different tier and some rates changed mid-month. Model Pricing rolls your trace into per-model cost, request, and average-cost-per-request so you can spot the model line where your assumed rate and your real rate diverge.

How it works

  1. 1

    Export the trace JSON for the period you want to validate.

  2. 2

    Paste it above; the model table updates live as you type.

  3. 3

    Compare avg-cost-per-request against your assumed rate; investigate any model flagged as a concentration anomaly.

Benefits

Per-model cost, share, and average cost per request in one table.
Flags any model holding ≥ 50% of spend so concentration risk is obvious.
Reveals rate drift — if avg-cost-per-request shifts, the assumed rate is wrong.
No vendor login, no sign-up — paste and read.

Frequently asked

Catch rate drift the day it happens

Pro saves per-model history and alerts when avg-cost-per-request shifts.

Create free account →

Like what you see? Get early access to the full Auditor

Join the waitlist to save audit sessions and turn trace data into attribution reports.

Join waitlist