What counts against a Claude Code usage limit?

Claude Code is Anthropic's coding tool, run in a terminal (the text window where you type commands) or inside a code editor. On a Pro, Max, Team or Enterprise subscription, its work is measured against a usage limit: the allowance of work the plan includes. The work is done in tokens, the pieces of text the model reads and writes, and Anthropic states the allowance as a multiple of another plan's allowance, with no token count attached. The allowance refills in two windows, one every five hours and one weekly, which the page on the five-hour window and the weekly window explains. This page answers the next question: which activity draws from that allowance, and why Claude Code draws from it faster than chat. It is part of the guide to Claude Code usage limits. Every fact is Anthropic's own statement, read on 4 October 2026.

One allowance, shared across every Claude product you use

Anthropic's support article on using Claude Code with a Pro or Max plan states the rule: "Both Pro and Max plans offer usage limits that are shared across Claude and Claude Code, meaning all activity in both tools counts against the same usage limits" [1]. The same article says the plan "also covers Claude Code in supported IDEs, including VS Code, Cursor and other VS Code forks, and JetBrains IDEs like IntelliJ and PyCharm", and that "IDE usage counts toward the same usage limits shared across Claude and Claude Code" [1]. An IDE is a code editor with built-in tools.

The Team and Enterprise article says the same for a seat, which is one member's place on the plan: "IDE usage is limited and billed the same way as terminal usage on your plan" [2]. Anthropic's Claude Code cost documentation adds Cowork, Anthropic's product for longer tasks outside coding: on Teams and Enterprise plans the per-seat allowance "is shared with Claude chat and Cowork" [4]. The pricing page puts it in one sentence for every plan: "Your activity across Claude on web, desktop, mobile, and Claude Code all draws from the same pool" [7].

The usage credits article, which covers paid use after the included limit, confirms the sharing holds there too: "Usage credits apply to both Claude conversations and Claude Code terminal usage. Your combined usage across both interfaces counts toward your limits" [8].

So one allowance covers the chat and the terminal. A long chat in the morning leaves less for the terminal in the afternoon, and the reverse.

Why a coding session uses more than a chat

Anthropic's Claude Enterprise consumption guide, read on 4 October 2026, ranks its own products by how many tokens they use. Its table gives chat as lower intensity, where "token usage scales with message length and conversation history", and Claude Code as higher intensity: each coding session "includes system prompts, file context, tool calls, and multi-turn reasoning", which Anthropic says means more tokens per session than chat [3]. Cowork is also higher intensity, because "agentic workflows, multi-step task execution, and Skills generate significant intermediate token usage that may not be visible to end users" [3]. The guide adds, in advice to administrators: "A single Cowork task or Claude Code debug session can consume many more tokens than chat" [3].

The Claude Code cost documentation gives the same reason in its own words: each Claude Code turn includes file contents, tool calls and multi-step reasoning, "so one debugging session can consume more than a day of chat" [4].

Each term in that list means something concrete:

• System prompts are the instructions Claude Code sends to the model with every request, before your own text.
• File context is the content of the files Claude reads, which travels to the model with the request.
• Tool calls are the actions Claude asks Claude Code to perform, such as reading a file or running a test. Each round of tool results is a further request, and Anthropic says Claude Code "sends your full conversation with every request" [4].
• Multi-turn reasoning is the model's reasoning across the several requests one task takes. Part of it is extended thinking, the model's working before each reply: Anthropic says extended thinking is on by default, that "thinking tokens are billed as output tokens", and that the default budget "can be tens of thousands of tokens per request depending on the model" [4].

Because the full conversation travels with every request, a session that has been open for hours draws more per message than a new one. Anthropic lists eight causes under the heading "Why usage climbs in a long session" [4]; the page on why usage rises in a long session takes each in turn.

Work Claude Code does without your typing counts too. A subagent is a second copy of Claude that works on one task in its own conversation; Anthropic says "the subagent's own requests still draw on your usage" [4]. Background jobs that summarise conversations for claude --resume, and commands such as /usage that check status, use tokens; Anthropic puts that background use at under $0.04 per session, typically [4]. The /insights report, which analyses your recent sessions, runs through the same account, and "the tokens count against your plan or API usage" [4]. Prompt suggestions, the short request Claude Code sends after each reply to suggest your next prompt, are skipped "while your account is close to or at its usage limit" [4].

What counts against the shared limit, activity by activity

Activity	Counts against the shared plan limit?	Source
Claude chat on web, desktop and mobile	Yes	Pro or Max article [1]; pricing page [7]
Claude Code in the terminal	Yes	Pro or Max article [1]
Claude Code in VS Code, Cursor, other VS Code forks, or JetBrains IDEs	Yes, the same limits	Pro or Max article [1]; Team article [2]
Cowork	Yes, on Teams and Enterprise seats by the cost documentation; the pricing page says all activity across Claude and Claude Code draws from one shared allowance	Cost documentation [4]; pricing page [7]
A subagent's own requests	Yes	Cost documentation [4]
The /insights report	Yes, against your plan or API usage	Cost documentation [4]
Background summarisation for claude --resume and status checks	Yes; Anthropic puts it at under $0.04 per session, typically	Cost documentation [4]
Research mode in the Claude app	Yes; usage credits apply once you pass the included limit, and Anthropic says research sessions may use tokens more quickly	Usage credits article [8]
Documents stored in project files	Yes, when used in a conversation, as part of the context	Usage credits article [8]
Requests sent with an ANTHROPIC_API_KEY set in your environment	No; they are billed as API usage to the Console account instead of the subscription	Pro or Max article [1]
Usage after the included limit with usage credits on	Billed separately at standard API rates, with the five-hour reset unchanged	Usage credits article [8]
A 529 Overloaded error from Anthropic's service	No; Anthropic says a 529 "is not your usage limit and doesn't count against your quota"	Error reference [5]
Server is temporarily limiting requests	No; Anthropic calls it a short-lived throttle unrelated to your plan quota	Error reference [5]

The API key row needs a plain warning, in Anthropic's words: "If you have an ANTHROPIC_API_KEY environment variable set on your system, Claude Code will use this API key for authentication instead of your Claude subscription (Pro, Max, Team, or Enterprise plans)", and the result is API usage charges in place of the subscription's included usage [1]. The error reference gives the check: run /status and confirm the active credential is the one you expect [5]. The page on a subscription or an API key for one developer compares the two ways of paying.

Model-specific limits, and when switching models helps

Anthropic's error reference lists four limit messages. Two cover every model and two cover one model family each:

You've hit your session limit · resets 3:45pm
You've hit your weekly limit · resets Mon 12:00am
You've hit your Opus limit · resets 3:45pm
You've hit your Sonnet limit · resets 3:45pm

The rule, in Anthropic's words: "The session and weekly limits are shared across all models, so switching models doesn't restore access. The Opus and Sonnet limits each apply only to requests to that model family, so switching to a model outside the family with /model keeps you working" [5]. /model is the command that changes which model the session uses. Anthropic adds the cost of that switch: "Each model has its own prompt cache, so the next request re-reads the whole conversation with no cache hits" [5]. The prompt cache is the stored copy of request text the service reuses at a lower price; the page on switching model or effort mid-task covers what one switch costs.

Two further messages are about entitlement, which is what a plan includes, and appear even with allowance left. Usage credits required for 1M context means the selected model uses the 1M-token extended context window, "and your plan only includes it through usage credits"; Anthropic says this "fires even when your session and weekly allowances have capacity remaining" [5]. The fix it gives is /model and the variant without the [1m] suffix, or /usage-credits to turn usage credits on, after which you restart Claude Code [5]. For the Fable models, Anthropic's pricing page lists Fable on the Pro plan as available through usage credits, and on Max 5x and Max 20x as "50% of weekly limits", with an asterisk on the two Max cells [7]; the error reference says Claude Code asks you to confirm before a Fable request bills usage credits, where your account requires that consent [5]. The page on usage credits after the limit covers both.

What the /usage breakdown tells you about your own usage

On a Pro, Max, Team or Enterprise plan, /usage shows a breakdown of what counts against the plan limits, by Anthropic's cost documentation read on 4 October 2026. It has three parts [4]:

• Attribution: recent usage attributed to skills, subagents, plugins and individual MCP servers, each as a percentage of the total. A skill is a packaged set of instructions for one task; a plugin is an installable bundle of such tools; an MCP server is a program that connects Claude Code to an outside tool or data source. Anthropic says an MCP server's share counts only the requests that consumed one of its tool results, and that before v2.1.222 Claude Code attributed every later request to a server after one call, overstating its share.
• Behavior flags: behaviours such as long context or cache misses, flagged when one accounts for 10 percent or more of recent usage, "each with a tip to reduce it".
• Loops: a row for each of the heaviest scheduled tasks that ran recently, with how often each fires and its total and per-run tokens. This part requires v2.1.242 or later.

Press d or w to switch between the last 24 hours and the last 7 days. Anthropic says the figures are computed from local session history on this machine, so usage from other devices or claude.ai is left out of the breakdown [4]. The breakdown is therefore the place to find out which of your own habits draws most from the allowance. The page on how to read /usage and /context explains each row.

Anthropic's own tips for staying inside a limit

The support articles give their own advice, read on 4 October 2026.

The Pro or Max article, under the heading "Staying within your plan", says to decline the API credit option when it is presented, to allow your usage period to reset before continuing, and to monitor your remaining allocation with /status [1].

The usage credits article lists five tips under "Tips for cost-effective usage". Choose efficient models: "Use our most efficient Haiku model, or the most recent Sonnet model for most tasks", which Anthropic says cost less per token than Opus. Use your plan's included usage first, and "plan intensive work sessions around your five-hour usage windows". Start new conversations for new topics "to minimize the size of your context window". Store documents you refer to often in project knowledge instead of re-uploading them. Start with conservative spending caps and adjust [8].

The consumption guide adds the effort level, the setting for how much the model thinks before it replies: "higher effort levels consume more tokens than lower ones", and Anthropic's advice to administrators is to reserve the highest level for the most demanding tasks [3]. The power user tips support article lists the levels as low, medium, high, xhigh, max and auto, and says the max level uses usage limits faster, so turn it on for one session at a time [6]. The page on which model and effort level to use goes through that choice.

The same power user article recommends running three to five Claude Code sessions at once, each in its own git worktree [6]. Each of those sessions sends its own requests, and the Pro or Max article says all activity counts against the same usage limits [1], so five sessions draw from one allowance five times as fast as one session doing the same work. That is our arithmetic on Anthropic's rule.

Where to start

Our position: the first thing to change is the length of the conversation, because the full conversation is sent with every request and the breakdown's long-context flag will tell you when that is 10 percent or more of your usage [4]. The second is the model and effort level, because the support articles name Sonnet and Haiku as the lower-cost choices [8] and the max effort level as the one that uses the allowance fastest [6]. Reveneau recommends checking the /usage breakdown before changing anything, so the change matches what the flags show.

Reveneau is an AI software development consultancy, and all of its code is written by AI, so the usage limit is a running constraint on every Reveneau build. Reveneau is independent of Anthropic. Every figure on this page is Anthropic's own statement about its own product, read on 4 October 2026.