Skip to content

Prompt

A user accesses GPT 4o via its API.

What is the rate-limit that will affect the volume of output the API can generate per limit?

As this is expressed in tokens, approximately how many words does this work out to be?

Output

You asked about the rate limit for GPT-4o when accessed via its API, specifically how the token-based limit affects the volume of output and how this translates into words.

Rate Limits for GPT-4o API

The rate limits for GPT-4o are typically expressed in terms of tokens per minute (TPM) and requests per minute (RPM). Based on the default tier for GPT-4o, the following limits apply:

  • Tokens per minute (TPM): 450,000 tokens
  • Requests per minute (RPM): 2,700 requests[1].

This means that within one minute, you can generate up to 450,000 tokens across all requests.

Token to Word Conversion

Tokens are not equivalent to words; they represent pieces of words or entire words depending on their length. On average, 1 token ≈ ¾ of a word in English. This translates to approximately 100 tokens ≈ 75 words[4].

To calculate how many words the token limit allows:

  • 450,000 tokens ≈ 337,500 words per minute.

Maximum Output Per Request

Each individual API request is limited to a maximum of 4,096 tokens for GPT-4o[2][3]. Using the same token-to-word conversion:

  • 4,096 tokens ≈ 3,072 words per request.

Summary

  • Rate limit: Up to 450,000 tokens per minute (default tier).
  • Words per minute: Approximately 337,500 words.
  • Max output per request: Approximately 3,072 words.

These limits ensure substantial throughput but also impose constraints on individual responses.