Count tokens for Claude models

This page shows you how to use the count-tokens endpoint to get the number of tokens in a message before you send it to a Claude model. You can use the token count to ensure your prompts don't exceed the model's context window.

There is no charge for using the count-tokens endpoint.

Supported Claude models

The following models support counting tokens:

Supported regions

The following regions support counting tokens:

  • us-east5
  • europe-west1
  • asia-east1
  • asia-southeast1
  • us-central1
  • europe-west4

Count tokens in basic messages

To count tokens, send a rawPredict request to the count-tokens endpoint. The request body must contain the model ID that you want to use.

REST

Before using any of the request data, make the following replacements:

  • LOCATION: A region that supports Anthropic Claude models. To use the global endpoint, see Specify the global endpoint.
  • MODEL: The model to count tokens against.
  • ROLE: The role associated with a message. You can specify a user or an assistant. The first message must use the user role. Claude models operate with alternating user and assistant turns. If the final message uses the assistant role, then the response content continues immediately from the content in that message. You can use this to constrain part of the model's response.
  • CONTENT: The content, such as text, of the user or assistant message.

HTTP method and URL: 

POST https://LOCATION-aiplatform.googleapis.com/v1/projects/PROJECT_ID/locations/LOCATION/publishers/anthropic/models/count-tokens:rawPredict

Request JSON body: 

{
  "model": "MODEL",
  "messages": [
    {
      "role": "user",
      "content":"how many tokens are in this request?"
    }
  ],
}

To send your request, choose one of these options:

curl

Save the request body in a file named request.json, and execute the following command: 

curl -X POST \
     -H "Authorization: Bearer $(gcloud auth print-access-token)" \
     -H "Content-Type: application/json; charset=utf-8" \
     -d @request.json \
     "https://LOCATION-aiplatform.googleapis.com/v1/projects/PROJECT_ID/locations/LOCATION/publishers/anthropic/models/count-tokens:rawPredict"

PowerShell

Save the request body in a file named request.json, and execute the following command: 

$cred = gcloud auth print-access-token
$headers = @{ "Authorization" = "Bearer $cred" }

Invoke-WebRequest `
    -Method POST `
    -Headers $headers `
    -ContentType: "application/json; charset=utf-8" `
    -InFile request.json `
    -Uri "https://LOCATION-aiplatform.googleapis.com/v1/projects/PROJECT_ID/locations/LOCATION/publishers/anthropic/models/count-tokens:rawPredict" | Select-Object -Expand Content

You should receive a JSON response similar to the following.

For information about how to count tokens in messages with tools, images, and PDFs, see Anthropic's documentation.

Quotas

By default, the quota for the count-tokens endpoint is 2,000 requests per minute.