How subscription usage limits work
An API bills every token; a subscription charges a flat monthly price and rations use instead. Anthropic's help centre describes two timers on Claude's paid plans: a session limit that resets every five hours and weekly limits that reset once a week. Use on claude.ai, Claude Code and Claude Desktop all counts toward the same allowance, so a heavy coding session leaves less for chat.
Higher plans buy a bigger allowance rather than a different product. Anthropic's pricing page sells Claude Max as 5x or 20x more usage per five-hour session than Pro, and states that Claude Code is included in all paid plans, drawing from the same pool as your chats.
What drains the allowance
Usage is not counted in messages. Anthropic says it depends on the length and complexity of conversations, the features used, the model chosen and the effort level selected. Tools and connectors are token-intensive, and long conversations that trigger automatic context management consume more of the limit. In token terms, every reply in a long chat re-reads the whole conversation, so the hundredth message costs far more allowance than the first.
That makes agents the fastest way to hit a limit. A coding agent that reads files, runs commands and resends a growing history on every step consumes in minutes what a chat uses in days. Our journal post on Claude Code caching shows how cache misses make it worse.
Subscription or API?
The subscription is cheaper when you use it heavily but within the limits, because the flat price stays flat. The API is cheaper for light or bursty use, and the only option when you need throughput the limits will not give you, or a guaranteed answer instead of a wait for the next window. The two are not linked by any published exchange rate — a plan's allowance is not stated in tokens — which is why our subscription vs API calculator compares them on your own workload.