GPT-5 Token Limits

GPT-5 token limits determine how much information you can provide to the model and how much content the model can return in a request.

What consumes the limit

For the original gpt-5 API model, OpenAI currently documents a 400,000-token context window and a 128,000-token maximum output.

These numbers are much larger than the typical prompts used in everyday applications, but developers still need to manage them carefully.

Leave room for a useful response

Your input can contain more than the text that a user has just typed. Depending on your architecture, the request may include system instructions, conversation history, retrieved documents, function or tool definitions, application metadata, and other supporting material.

For example, imagine a chatbot where each conversation adds thousands of tokens. If the complete history is repeatedly sent to the API, the request becomes larger as the conversation continues. Eventually, older messages may need to be removed, summarized, or stored outside the active context.

Check the model-specific constraint

A common solution is context management. Applications can summarize previous messages, retrieve only relevant knowledge, and retain important user state separately rather than indefinitely sending every previous message.

Token limits should also not be confused with API rate limits. A context limit controls the size of an individual model interaction, while rate limits control how many requests or tokens your account can process over time.

Practical checks before you proceed

Knowing the difference makes it easier to diagnose errors and design scalable AI applications.