AI spend you can forecast

Seats on personal cards, API invoices, expense reports — Kordon rolls them into one report by team, with budgets and a forecast

Before: subscriptions on personal cards, API invoices, reseller contracts and expense reports. After: one report by team — Engineering $2,680, Sales $1,480, Support $995, Legal $655, Marketing $590; total $6,400.

Where the money goes

Complex tasks are few, but they take most of the bill. Kordon sends the simple ones to cheaper models

Simple tasks58% of volume
Routine tasks33% of volume
Complex tasks9% of volume
Gemini 3.5 Flash-Lite$240
Llama on your servers$0
Gemini 3.1 Pro$920
Claude Opus 5.5$1,440
Simple tasks
58% of volume — 10% of the bill
Complex tasks
9% of volume — 55% of the bill
Local model
Confidential tasks — $0 in token fees

Monthly total on the pay-as-you-go API: $2,600.

Three ways to pay for models

Each has its own price and risk. Kordon works with any of them — or several at once

flexible but pricier

Provider APIs

You pay for every request at the model's rate. Any model is available, but the bill grows with usage.

Price
high
Ban risk
low
Model choice
any
predictable

Seat per employee

A fixed price per employee, within the plan's usage limits. A banned account affects one person.

Price
medium
Ban risk
low
Model choice
narrow
cheap but risky

Shared seat pool

A few accounts for the whole company — the cheapest option. But OpenAI and Anthropic terms prohibit account sharing: one ban cuts everyone off.

Price
low
Ban risk
high
Model choice
medium

Run your own numbers

Strongest model
How you pay

Our assumptions: 20% of employees use AI constantly, 50% regularly, 30% occasionally. Kordon sends simple requests to a light model and complex ones to the strongest.

Monthly model bill · pay-as-you-go API$3,572$42.9K a year · $24 per employee
  • Provider APIs$3.6K
  • Seat per employee$5.8K
  • Shared pool$1.4K

Without Kordon, sending everything to GPT-6 Astra would cost $10,416 a month. Routing by complexity saves $6.8K.

List prices as of October 2026, USD, before tax. Kordon license not included.

Budgets without surprises

Each team spends within its own budget — the gateway enforces it

  • A dollar budget per key, team or project
  • At 80%, the manager gets an alert
  • At 100%, requests stop until the budget is raised
  • The monthly bill never exceeds the budget
AI budgets · Octoberday 30 of 30
  • Engineering$6.5K of $7.5K

    80% of budget: the manager has been alerted

  • Support$2.8K of $4K

  • Marketing$1.9K of $2K

    80% of budget: the manager has been alerted

  • Delta (contractor)$600 of $600

    Budget spent: new requests get a 429

Estimate savings on your AI traffic

An engineer reviews how you connect to models and sets up a test environment in your network — before you buy

Talk to us