How the web, desktop and command-line entry points differ, how Free / Pro / Max / Team / Enterprise are split, and how the two usage windows — five hours and seven days — are counted. Pick the tier that matches what you actually do, and spend the allowance where it matters.
Web, desktop, command line — each covers a stretch
Entry point 01
claude.ai on the web
The main entry point: nothing to install, usable anywhere. You reach claude.ai through your dedicated usage environment and out of your own residential egress, and conversations, document and image analysis, Projects, web search and Artifacts all work as they do officially.
The first choice for everyday questions, writing, analysis and team work
Used through your dedicated usage environment; requests leave from your own residential egress
Team and Enterprise manage seats and review usage in the web interface
Entry point 02
The desktop app
Claude Desktop is a native local application with the same experience as the web, and it makes local MCP connectors and system-level shortcuts easier to reach. It goes out of the same residential egress and draws on the same allowance as the web, so avoid running both hard at once.
Entry point 03
The Claude Code CLI
An agentic coding agent in the terminal: it reads and writes files in a repository, runs commands and carries out multi-step work. It comes with the Pro / Max / Team / Enterprise subscriptions (not with Free), and its usage shares one counter with your conversations.
Every entry point, one egress
Web, desktop or command line, requests end up leaving for claude.ai from your own residential egress. The access pattern stays consistent — do not switch egress or run from several places at once.
One allowance across entry points
Within one plan, web conversations, Claude Code and Cowork share a single usage counter — whichever tool you reach for, it draws on the same pool. Review it in one place under Settings > Usage.
One path
Three entry points, one egress
Web, desktop and command line all go through your dedicated usage environment and leave for claude.ai from your own residential egress.
The egress is fixed and does not drift, so access keeps looking like the same residential connection it always was.
On the command line, TUN takes over outbound traffic so CLI requests travel the same path instead of slipping past the local configuration.
Three entry pointsWeb / desktop / command line
Your residential egressDedicated · fixed
claude.aiUpstream
Plan comparison
Pick the tier your usage actually needs
Individual · trialFreeFree of charge
Try conversations, document and image analysis first; includes the current Sonnet, not Opus and not Claude Code.
Individual · dailyProAbout $20 / month
Heavy daily individual use and light-to-moderate development; includes Claude Code, with several times the Free allowance.
Most popularIndividual · heavyMax 5xAbout $100 / month
Around 5× the Pro usage per session; for professionals who run Claude Code daily and would rather not keep hitting the limit.
Individual · all dayMax 20xAbout $200 / month
Around 20× the Pro usage per session; for people who treat Claude and Claude Code as their main tool nearly all day.
Team · collaborationTeamFrom about $25 / seat
Five seats minimum, with central billing and collaboration; a premium tier at roughly $125 / seat carries a higher allowance.
Organisation · governanceEnterpriseCustom / on request
SSO, access governance, compliance and a negotiable allowance; priced per seat, arranged through sales.
The Claude plans compared on available models, usage windows, Claude Code, who they suit and price range
Plan
FreeFree of charge
ProAbout $20 / month
Max 5xAbout $100 / month
Max 20xAbout $200 / month
TeamFrom about $25 / seat
EnterpriseCustom / on request
Models available
The current Sonnet
Sonnet + Opus + Haiku
As Pro, with early access
As Pro, with early access
As Pro + collaboration admin
The full current line-up
5-hour rolling window
Basic cap, clearly lower
The baseline allowance
About 5× Pro
About 20× Pro
Scales with seats
Negotiated
7-day weekly window
Basic cap
All models + Sonnet-only, two caps
All models + Sonnet-only, two caps
All models + Sonnet-only, two caps
All models + Sonnet-only, two caps
Negotiated
Claude Code
Not included
Included
Included
Included
Included
Included
Who it suits
First look, light use
Heavy daily individuals, light to moderate developers
Professionals running Claude Code daily
All-day agentic coding or research
Small teams and departments of 5+
Large organisations that need governance
Price range
Free of charge
About $20 / month
About $100 / month
About $200 / month
From about $25 / seat
Custom / on request
Prices and annual discounts change, and Claude's own pricing page is the authority; paid tiers are quoted in US dollars here and not converted. The model row says "current" on purpose: today that is Opus 4.8 / Sonnet 4.6 / Haiku 4.5 (with the stronger Fable 5 alongside), and what is actually open follows the official line-up. Team tiers and prices are reported inconsistently (standard around $25 / seat, premium around $125–150 / seat including Claude Code, with a $30 figure also in circulation) — take the official page as final. "About 5×" and "about 20×" are the official qualitative statements about usage per session; the message counts that circulate online (around 225 or around 900 messages per 5 hours) are third-party estimates rather than an official commitment, and real usage moves with message length, attachments, tool calls and the model you pick. Whether the weekly cap resets separately for Opus is shown in your own account under Settings > Usage.
Usage management
Spend the window where it counts
How the windows are counted
The 5-hour rolling window starts from your first message and resets roughly five hours later — it is not pinned to the clock.
The 7-day weekly window resets after about seven days, and there are two of them at once (all models, and Sonnet only); the stronger the model, the faster it eats its own weekly cap.
Usage shares one counter across web, desktop, Claude Code and Cowork; since 2026-05-06 the 5-hour Claude Code limit is doubled for Pro / Max / Team / per-seat Enterprise.
How not to waste it
Plan ahead and state the whole task once: fold related questions into a single message instead of clarifying back and forth.
Park stable context in a Project (it is cached, so re-use only counts the new part) and use memory to refer back to earlier conclusions.
Choose the model per task: Sonnet or Haiku for everyday work, Opus for the jobs that genuinely need the reasoning.
When to move up a tier
If you keep hitting the 5-hour or weekly limit (especially running Claude Code hard), move up the ladder: Pro → Max 5x → Max 20x.
Team for shared work and central billing; Enterprise when you need governance, compliance and a custom allowance.
Watch the two progress bars under Settings > Usage, and ease off or switch to a cheaper model as you approach the cap.
5-hour rolling windowCounted from your first message
The whole window resets after five hours, however many messages you sent.
7-day weekly windowAll models + Sonnet only
The stronger the model, the faster it eats its own weekly cap; it resets after about seven days.
Spent
Current
Available
When you hit the limit, pace yourself
Near the cap, slow down, switch to a cheaper model, or wait for the window to reset — heavy work goes furthest right after a reset. Do not open extra accounts or move to another egress to get more allowance: that breaks the one-person-per-seat rule and raises the risk of account controls.
Common questions
About usage and plans
Can Free use Claude Code?
No. Claude Code, the terminal coding agent, comes with the Pro / Max / Team / Enterprise subscriptions and is not part of the Free tier. Free can hold conversations and analyse documents and images in the browser with the current Sonnet, but has no Opus-class model and no command-line coding agent.
Are "around 225" and "around 900" messages official numbers?
No. The only official statement is qualitative — around 5× or 20× the Pro usage per session. The specific message counts that circulate are third-party estimates, not an official commitment. What you actually get moves with message length, attachment size, accumulated conversation length, tool calls and the model you choose, so take your own account's Settings > Usage page as the authority.
How are the 5-hour and weekly windows counted?
The 5-hour window starts from your first message and resets as a whole about five hours later (not on a fixed clock). The weekly window resets after about seven days, and two of them run at once — all models, and Sonnet only — with the stronger models reaching their own weekly cap faster. Both windows apply at the same time, so heavy work goes furthest right after a reset.
Do web, desktop and command line have separate allowances?
No, they are not separate. Within one plan, Claude on the web and desktop, Claude Code and Cowork share a single usage counter — whichever tool you use draws on the same pool. So opening more entry points does not accumulate more allowance; review and plan it in one place under Settings > Usage.
Pro, Max 5x or Max 20x?
It comes down to how often you hit the limit. One person, budget-conscious, light to moderate coding: start with Pro. Running Claude Code every day and tired of hitting the cap: Max 5x. Near-continuous heavy agentic coding or research, with $200 a month acceptable: Max 20x. For shared work and central billing, Team; for governance, compliance and a custom allowance, Enterprise. Prices change — Claude's own pricing page is the authority.
