DORETPRISM

GPT-5.5 cost per customer calculator

What GPT-5.5 costs you for each customer a month, at OpenAI's list prices ($5.00 input, $30.00 output per 1M tokens), next to every other model. It updates as you type.

Per 1M tokens: $5.00 input, $0.50 cached input, $30.00 output. Above 272K input tokens: $10.00, $1.00, $45.00.
What one request is
Prompt, history, documents and tool results sent with it.
Reasoning models bill their thinking as output: include it.
%
A long system prompt or a conversation that repeats is often 50–90% cache hits. Cache writes are not counted.
$
AI cost a month
$1,320.00
$15,840.00 a year
Per customer, on average
$6.60
$0.0220 per request
To keep AI under % of revenue, a customer has to pay at least $33.00 a month on average.

The average hides your expensive customers

Usage is rarely even. An illustration, assuming % of customers make % of the requests:

A heavy customer
$33.00
5× the average
Everyone else
$3.67
per customer

An assumption, not your data: the real split shows only in your usage.

DORET Prism reads the usage reports OpenAI and Anthropic already keep and shows spend by customer, for every customer with its own API key or project (Startup plan and up, and the launch trial). No code changes.

The same workload on every model

200 customers × 300 requests a month, 2,000 input and 400 output tokens each. Pick a row to use that model above.

ModelPer requestPer customerA monthvs GPT-5.5
$0.0004$0.12$24.000×
$0.0004$0.12$24.000×
$0.0005$0.16$32.400×
$0.0013$0.39$78.000.1×
$0.0033$0.99$198.000.2×
$0.0040$1.20$240.000.2×
$0.0040$1.20$240.000.2×
$0.0072$2.16$432.000.3×
$0.0080$2.40$480.000.4×
$0.0080$2.40$480.000.4×
$0.0120$3.60$720.000.5×
$0.0160$4.80$960.000.7×
$0.0220$6.60$1,320.00selected
$0.0320$9.60$1,920.001.5×
$0.0400$12.00$2,400.001.8×

Price is one side of the choice: models differ in quality and speed for your task, so test before you switch.

Standard prices as of November 14, 2026, from OpenAI and Anthropic (rates used: $5.00 input, $0.50 cached, $30.00 output per 1M tokens for GPT-5.5). Batch, fast mode, data-residency endpoints and cache writes change the real price and are not counted.

How the numbers are worked out

The formula

One request costs its uncached input tokens at the input price, its cached input tokens at the cache price and its output tokens at the output price, each per million tokens. That times requests per customer is the cost per customer; times customers, the month. When a request's input passes a model's long-context threshold, the whole request is priced at the long-context rates, as the providers do.

What does GPT-5.5 cost per million tokens?

$5.00 for input, $0.50 for cached input and $30.00 for output, at OpenAI's standard rates; above 272K input tokens a request costs $10.00, $1.00 and $45.00. Prices as of 2026-11-14, from OpenAI's pricing page.

How many tokens is a request?

In English a token is roughly four characters, or three quarters of a word. Input is everything sent with the request: the system prompt, the conversation so far, documents and tool results. That is why an agent that calls tools several times reads far more than it writes. The three presets are starting points; your provider's usage page shows your real averages.

Why do some customers cost so much more?

Because usage is uneven: a few customers send most of the requests, keep longer conversations or attach bigger documents. An average per customer hides them, and a flat price per seat can lose money on exactly those customers.

What is not counted

Batch discounts, fast mode, data-residency surcharges and the cost of writing to the prompt cache. Taxes and negotiated rates are not included either.

How do I see my real cost per customer?

DORET Prism reads the usage reports OpenAI and Anthropic already keep and splits spend by customer, for each customer that has its own API key or project. It needs an admin key and no code changes; here is which key and how to revoke it.