Track spend

Proxium records the cost of each call that a model answered. You read the cost in the console, or from the API.

The answer of the modelusage: 1,200 tokens in, 300 out

→↓

× the price of that modelfrom Routing › Model catalog

→↓

The cost of the call$0.00036

→↓

Where you read itOverview, Requests, GET /usage/me

Where to read the spend

You want to know	Where
The spend of today, 7 days or 30 days	Console: Overview › Spent
The spend of each application	Console: Overview › Per-app burn
The spend of each virtual key	Console: Overview › Per-key burn
The spend of each model	Console: Overview › Model split. API: by_model of /usage/me
The spend of each task	API: by_task of /usage/me
What the cache saved	Console: Overview › Saved by cache, and Requests › Cache
The spend of this month, from your code	API: GET /usage/me

Each console screen has a Time window: today, 7d or 30d.

How Proxium computes the cost

Proxium reads the token counts from the usage block of the answer. It multiplies each count by the price of the model that answered.

Example: a call to a model at $0.15 for each million input tokens and $0.60 for each million output tokens.

Tokens	Count	Price for each million	Cost
Input	1,200	$0.15	$0.00018
Output	300	$0.60	$0.00018
The call			$0.00036

Two more token types have their own price:

Tokens	Price
Prompt-cache read	The cache-read price of the model. If it has none, the input price
Prompt-cache write	The cache-write price of the model. If it has none, the input price

A vendor counts its cache-read tokens inside the input total. Proxium takes them out of the input count first, so it prices each token one time.

To see the price of a model, open Routing › Model catalog, column Cost, per Mtok.

Cases that change the cost

Case	Recorded cost
The model has no price, or a price of $0	$0.00. A new model that Proxium finds at a vendor starts at $0
A stream with no usage block	Proxium asks the vendor for one with stream_options.include_usage. If none arrives, it counts 4 characters as 1 token
A second model answered after a failure	The cost of the model that answered
A cache hit	$0.00 by default. See What a hit costs

Name each application

To see the spend of each application, send the x-proxium-source header with the name of the application. Set it one time, on the client:

Python

client = OpenAI(
    base_url="https://proxium.tech/v1",
    api_key=os.environ["PROXIUM_KEY"],
    default_headers={"x-proxium-source": "support-bot"},
)

Node

const client = new OpenAI({
  baseURL: "https://proxium.tech/v1",
  apiKey: process.env.PROXIUM_KEY,
  defaultHeaders: { "x-proxium-source": "support-bot" },
});

curl

Terminal

curl https://proxium.tech/v1/chat/completions \
  -H "Authorization: Bearer $PROXIUM_KEY" \
  -H "Content-Type: application/json" \
  -H "x-proxium-source: support-bot" \
  -d "{\"model\": \"$PROXIUM_MODEL\", \"messages\": [{\"role\": \"user\", \"content\": \"hello\"}]}"

A call with no header counts under the name of the project.

The name also selects the ceiling of the application, and its routing rule. See Set budgets and limits and Route requests to models.

To group calls by a task of your own, also send x-proxium-task-id, for example ticket-4812. /usage/me then gives the spend of each task in by_task.

tip

Give each virtual key the name of its application when you create it. Per-key burn shows a key without a name as unnamed.

Read the spend from the API

GET https://proxium.tech/usage/me gives the spend of the project in the current calendar month. The path has no /v1.

Terminal

curl https://proxium.tech/usage/me \
  -H "Authorization: Bearer $PROXIUM_KEY"

Answer (shortened)

{
  "tenant_slug": "acme",
  "period": "month",
  "credits": { "unlimited": true, "used_millicents": 184250, "used_usd": 1.8425 },
  "requests": 412,
  "tokens": 903114,
  "by_model": [
    { "provider": "t.acme.groq", "model": "llama-3.3-70b-versatile", "cost_usd_millicents": 184250, "tokens": 903114, "requests": 412 }
  ],
  "by_task": [
    { "task_id": "ticket-4812", "cost_usd_millicents": 2210, "tokens": 10840, "requests": 3 }
  ]
}

Field	Holds
credits.used_usd	The spend of this month, in dollars
credits.used_millicents	The same spend, in millicents. 100,000 millicents are $1
credits.unlimited	true: the project has no monthly cap. During the open beta it is always true
requests, tokens	The calls and the tokens of this month
by_model	The spend of each model, highest first
by_task	The spend of each x-proxium-task-id, highest first. At most 100 rows

/usage/me has no split by application. For that, use Per-app burn in the console. Read the usage of the project gives every field.

Related pages

• Set budgets and limits: cap the spend of one application.
• The response cache: what a cache hit costs.
• The request flow: when Proxium records the cost.