The financial controller for AI inference
Find out what your AI inference is actually costing you.
InfFyn shows your spend per 1M tokens by model and finds where you're overspending — free.
No credit card. Read-only access or CSV upload. Takes about 5 minutes.
GPp1M by model — last 30 days
acme-prod
gpt-4o
1.42M requests
+$184
claude-3.5-sonnet
920K requests
+$62
gpt-4o (classify)
610K requests
−$41
gemini-1.5-pro
402K requests
+$22
gpt-4o (router)
248K requests
−$118
Blended cost view would show
$0.62 / 1K tokens
Illustrative data — your audit uses your real usage.
GPp1M
Cost tells you what you paid.
GPp1M tells you what you kept.
Gross Profit per 1M Tokens — profit per million tokens, broken out by model and feature. Not a blended average that hides the segments where you're losing money.
Your free audit starts with spend per 1M tokens. Add your revenue and it becomes GPp1M — profit per segment.
What you see today
company-wide average
Blended cost
$0.62 / 1K tokens
One flat number. Looks fine. Tells you nothing about which features pay for themselves and which quietly subsidize the rest.
Illustrative data — your audit uses your real usage.
What InfFyn shows you
range across segments
GPp1M, broken out
+$184 → −$118
Chat (gpt-4o)
+$184
Summarize (claude)
+$62
Classify on gpt-4o
−$41
Router on gpt-4o
−$118
Illustrative data — your audit uses your real usage. The two negative segments would be invisible in the blended view.
01
Profit per token, not just cost
Every request joined to the revenue event it produced. GPp1M by model, feature, customer cohort — so you can see where margin actually lives.
02
Find the waste
Mis-routed spend. Easy queries running on expensive models. Retries that nobody noticed. Ranked by dollar impact — the biggest leak first.
03
Continuous monitoring
Inference spend drifts week to week as traffic and prompts change. InfFyn re-runs your audit on a schedule and alerts you when GPp1M drifts on any segment.
How it works
Three steps. Roughly five minutes.
01
Connect or upload
Read-only API access, or upload a usage CSV — no credentials required.
02
See your spend per 1M tokens
Spend per 1M tokens by model, plus one quantified waste finding. Free.
03
Monitor continuously
Turn the snapshot into ongoing margin monitoring with drift alerts.
Read-only access
We never get write permissions to your AI accounts.
No raw-log storage
We compute on usage metadata, not the contents of prompts.
Upload option
If you'd rather not connect anything, send a usage CSV.
Find money you didn't know you were losing.
A free audit takes about five minutes. You'll see your spend per 1M tokens by model and one quantified waste finding.