Sign in / Join

Claude Code usage limits Why your agent stops mid-task.

Coding agents burn through usage faster than any chat. Here are the limits you run into with Claude Code, why agents hit them first, and the ways around the meter.

The three limits you meet

01

Usage windows

Subscription plans count usage over a rolling window of hours. When the window fills, the agent waits until it resets.

The catch: a long refactor can use a whole window in one go.

02

Weekly caps

Some plans add a ceiling across the week on top of the windows, especially for the largest models.

The catch: a heavy Monday can decide your Friday.

03

Per-token API billing

With an API key there is no window, but every input and output token is billed.

The catch: agent loops make the bill hard to predict.

The exact amounts depend on your plan and change over time. Check your provider’s current terms.

Why agents hit them first

Context grows every turn

Each step resends the conversation, the files it read and the tool output. Late steps cost far more than early ones.

Tool loops multiply calls

Run the tests, read the error, edit, run again. One task can be dozens of model calls.

Subagents run in parallel

Helpers that search and summarise in parallel use usage of their own at the same time.

Ways around the meter

Wait for the reset

Cost
Nothing extra
Downside
Work stops until the window resets

Pay per token

Cost
Grows with every loop
Downside
Hard to predict for agent runs

Flat time passThis is us

Cost
One price per day or three
Downside
Open-weight models; 2 sessions per key

FAQ

Does endlesstokens remove Anthropic’s limits?

No. It is a different backend: Claude Code talks to open-weight models we host, and during your pass we do not count or cap tokens or requests.

Is a time pass cheaper than paying per token?

It depends on how much you use. The heavier your agent runs in a day, the more a flat pass saves.

What are the endlesstokens limits?

Two parallel sessions per key, a 128K context window, up to 5 images per request and reasoning effort up to medium. If every GPU is busy, a new session can be briefly refused and the time is credited back.

Where do I find my current Claude Code limits?

In your Anthropic account and plan terms. The amounts depend on the plan and change over time.

Stop watching the meter.