Claude Code · Plans
Claude Code Subscription: Which Plan Do You Actually Need?
Prices vary between loads. We captured $18 annual / $22 monthly, Max $110, Ultra $220 at 11:23 (UTC+8), and the same page served $17 / $20, Max $100, Ultra $200 during a later check the same day. Both sets were rendered by Anthropic, not by third parties, so we do not pick a winner: the table below quotes the first capture and the figures may differ on the checkout screen.
The four plans, side by side
| Plan | Monthly | Usage | API credits | Claude Code |
|---|---|---|---|---|
| Free | $0 | Limited usage every 5 hours and every week | — | No |
| Pro | $22 $18/mo if billed yearly ($220 up front) | More usage than Free, same 5-hour + weekly windows | — | Yes |
| Max | $110 | 5× Pro usage per 5-hour window and per week | $100/mo | Yes |
| Ultra | $220 | 10× Pro usage per week; no 5-hour limit (verified accounts) | $200/mo | Yes |
The line most people miss: Anthropic's own FAQ says Claude Code “is included in all paid plans” and “shares the same usage limits as the rest of your plan.” So every chat message you send in claude.ai, and every agent task you run in Claude Code, draws from the same bucket.
Pick in three questions
- Do you already pay for Claude? No → start on Pro. You cannot use Claude Code on Free at all, so “try it for free” is not an option.
- Do you hit the limit before lunch? If you exhaust the 5-hour window on a normal working day, Pro is the wrong plan — not because the model is worse, but because the window resets on a schedule you don't control.
- Are you building a product on top? If you also need API access for your own scripts, the $100/$200 monthly API credits bundled with Max/Ultra change the maths, since API usage is billed separately from chat usage.
What actually costs extra
- Heavy agent runs. Claude Code reads files, runs commands and re-prompts itself, so in our experience one agent task consumes far more of the plan than a single chat reply does. Plan limits count that work.
- API usage. If you use your own API key inside Claude Code or in your own tools, that is metered API billing, not your subscription.
- Extra usage credits. Anthropic now lets paid plans turn on usage credits billed at standard API rates; treating the plan as unlimited is the fastest way to a surprise bill.
Three mistakes that waste money
1. Buying Max on day one
Max is 5× Pro. If you have not yet hit Pro's ceiling for a week straight, you are paying for headroom you cannot measure.
2. Assuming “5-hour window” means “5 hours of work”
The window is a rolling reset, not a stopwatch. Long agent sessions burn the window faster than short ones, so two identical days can end very differently.
3. Paying for a plan to get API access
Subscription and API are separate products with separate billing. If your goal is to call the API from your own code, price the API directly.
Questions people actually ask
Does Claude Code have its own subscription?
No. Claude Code is a client that uses your Claude plan. There is no separate Claude Code price — you buy a Claude plan, and the Code portion draws from the same limits.
How much does a Claude Code membership cost?
The cheapest path is Claude Pro at $22/month ($18/month if billed annually). Above it are Max at $110 and Ultra at $220. Prices as published on claude.com/pricing, checked 2026-11-07.
Is Claude Code free?
Not usable on the Free plan. Every paid plan includes it, so the practical minimum is Pro.
What is the cheapest plan that works for real coding?
Pro, if your sessions are short and you can spread work across the day. Max, if you run long agent tasks — the 5× allowance is what stops you from stopping mid-refactor.
So what should you do?
Start on Pro for one week. Track two numbers: how often you hit the 5-hour limit, and how many agent tasks you abandoned because of it. Then do the arithmetic on your own numbers — at the captured prices the gap between Pro and Max is about $90/month, so Max only pays for itself if being blocked costs you more than roughly $22 per blocked session across four or five sessions. If you never get blocked, ignore the upgrade page.
What each Claude Code subscription actually gets you
It helps to stop thinking of “Claude Code subscription” as one product. There is only the Claude plan, and Claude Code is the terminal agent that draws from it. What changes between plans is not features but how long you can work before the window closes.
Free — you cannot use Claude Code at all
The Free plan is a way to evaluate the chat product, not the agent. If your goal is a Claude Code subscription, Free is not a cheaper tier of it; it is outside the product entirely. This trips people up because Claude Code's marketing pages are reachable without a plan, and the failure only appears at the first prompt.
Pro — the evaluation plan
Pro is the cheapest plan that includes Claude Code. It is enough to install the CLI, wire up a repository, and run real agent tasks for a day. What it is not enough for is a long autonomous session: a single refactor can consume the same window that a week of chatting would have used, because the agent reads files, runs commands and re-prompts itself.
Max — the working plan
Max exists for people whose day is mostly agent work. The 5× allowance is the difference between “I stopped because I hit the limit” and “I stopped because I was done”. If you bill your hours, this is usually the first plan where the subscription cost is smaller than the cost of waiting.
Ultra — only for teams of one who act like a team
Ultra multiplies the weekly allowance again and bundles a larger monthly API credit. It is rational when you are running several agents in parallel, or when the same account also pays for API usage in your own tooling. For a single developer doing focused work, it is usually headroom you never measure.
Claude Code subscription vs paying for the API
These are two different purchases and the confusion is common. A Claude Code subscription is a flat monthly plan with usage windows; API billing is metered per token with no windows at all. If you write code that calls the API from your own scripts, your subscription does not cover it — and conversely, buying API credits does not give you a claude.ai plan.
Rough rule: if a human sits in front of the terminal prompting an agent, the subscription is almost always cheaper. If a machine loops through prompts with no human in the loop, price the API first — usage windows were designed for human-paced work, not for batch jobs.
How this Claude Code subscription comparison was compiled
We did not take the numbers from a press article. We opened claude.com/pricing, captured the plan cards, and recorded the FAQ text about which plans include Claude Code. The plan-card capture is on this page so you can check our work rather than trust it; the FAQ wording is quoted as text with the link above, and we have not screenshotted it.
What we have not done is measure how many agent tasks each tier survives. Anthropic does not publish a task-per-window figure. We have not found a sourced one, so treat any specific number you read elsewhere as an estimate until it links to a measurement method. The honest version: the 5× and 10× multipliers are the only numbers we can compare, and they are relative to Pro, not absolute.
Next
- coming Claude Code pricing: what a month really costs — the same numbers, with a cost-per-task model.
- coming Claude Code API key — when API billing beats a subscription.
- “Codex could not start” — if you are comparing agents and Codex is failing on you.