Claude 4 Token Limits

“Claude 4” refers to a generation of Anthropic models rather than one permanent token limit. Exact limits therefore depend on the specific Claude model and API configuration being used.

What consumes the limit

This distinction has become especially important as Anthropic has continued releasing newer Claude models. For example, Claude Sonnet 4.6 introduced a 1 million token context window in API beta, while Anthropic has subsequently introduced newer Sonnet models as well.

A large context window allows developers to include long documents, code, conversation history, business data, or multiple source files in a single workflow.

Leave room for a useful response

However, maximum context should not be treated as a target.

Sending hundreds of thousands of tokens on every request can dramatically increase usage and may add information that is irrelevant to the current task. Retrieval, summarization, and caching are often more efficient than repeatedly supplying entire datasets.

Check the model-specific constraint

Developers should also differentiate four concepts:

Input tokens are the tokens sent to Claude.

Practical checks before you proceed

Output tokens are the tokens Claude generates.

Context window represents the amount of information that can be available during the interaction.

Practical checks before you proceed

Rate limits control API throughput over a period of time.

Because Anthropic's model lineup and limits evolve, applications should store model configuration rather than hard-coding assumptions such as “Claude always supports X tokens.”

Practical checks before you proceed

If you specifically require very long context, check the documentation for the exact Claude model ID you're deploying and confirm whether extended-context access has additional conditions or pricing.