這 Claude Code Pro price is $20 a month, or $200 a year paid up front (about $17 a month). That part takes ten seconds. The harder question is the one nobody asks before they pay: which wall do you hit first? Pro gives you real Claude Code, Opus included, in your terminal and your IDE. It also gives you a 5-hour usage window, a weekly cap, and a few Claude Code defaults that only apply on Pro. Those decide whether $20 is a bargain or a trap.
I run a fleet of scheduled Claude Code agents for this site. They write the posts, check the inbox, and schedule the social posts. That fleet is 不是 on Pro, and I’ll show you exactly why, with the schedule and the numbers. Then I’ll give you a test you can run on your own account to see whether Pro, Max, Ultra or a plain API key is the cheapest option for how 你 work.
Every price below was re-checked on claude.com/pricing and Anthropic’s own docs on 22 November 2026. Prices exclude tax and Anthropic changes them, so check the date before you quote me.
Claude Code Pro Price: The 30-Second Answer (November 2026)

Here’s the whole lineup for individuals, straight from the pricing page:
| 計劃 | 價格 | Claude Code? | Usage vs Pro | 帳單 |
|---|---|---|---|---|
| 自由的 | $0 | 不 | “Limited” | 不適用 |
| Pro | $20/mo, or $200/yr up front (~$17/mo) | 是的 | 1x (5-hour + weekly) | Monthly or annual |
| 最大限度 | $100/mo | 是的 | 5x every 5 hours and every week | Monthly only |
| Ultra | $200/mo | 是的 | 10x weekly, no 5-hour limit (verified accounts) | Monthly only |
| Team standard seat | $20/seat annual, $25 monthly | 是的 | “More usage than Pro” | Monthly or annual |
| Team premium seat | $100/seat annual, $125 monthly | 是的 | 5x a standard seat | Monthly or annual |
Three things in that table matter more than the numbers:
- Pro is the cheapest way into Claude Code. The Free plan doesn’t include it. If you’re hunting for a $0 route, I covered every real option in is Claude Code free? the honest pricing breakdown. This page is only about the paid plan.
- Annual billing only exists on Pro (and Team). Max and Ultra are monthly only. The $200-a-year Pro deal saves $40 against twelve monthly payments. That’s worth it only if you’re sure you won’t outgrow Pro in the first few months, and plenty of people do.
- Claude and Claude Code share one usage pool. Anthropic’s support article is explicit: chats on claude.ai and work in Claude Code count against the same limits. Long research chats in the afternoon leave less room for coding in the evening.
So the Claude Code Pro price is simple. What you actually get for it, and where it stops, is the rest of this post.
What the $20 Actually Unlocks Inside Claude Code

Most pricing pages list the claude.ai features (Projects, Research, Claude in Chrome) and stop there. If you’re paying for Pro mainly for Claude Code, these are the parts that matter:
Models: Opus yes, Fable on usage credits
Pro includes Opus, Sonnet and Haiku. The pricing page’s model table marks Opus as “Yes” on Pro and “No” on Free. Fable, Anthropic’s top model, is the exception. On Pro it’s “paid with usage credits”, which means pay-as-you-go on top of the subscription. On Max and Ultra, Fable is included up to half your weekly limit. If you specifically want Fable for long agentic coding sessions, Pro’s sticker price isn’t really $20.
Dynamic workflows: off until you turn them on, and sized “small”
This is the Pro-only detail I didn’t see on any of the five pricing pages I read for this post. Claude Code’s dynamic workflows run many subagents in parallel from one prompt. The docs say they’re on every paid plan, but on Pro you have to switch them on yourself from the Dynamic workflows row in /config. Once they’re on, the default size guideline on Pro is 小的 (Claude aims for fewer than 5 agents) from Claude Code v2.1.271 onward. Everyone else defaults to 中等的 (fewer than 10).
That’s a sensible default, not a punishment. A workflow fanning out 30 agents would burn a Pro 5-hour window in minutes. But it tells you how Anthropic sees Pro: one person, one main session, a few helpers. If you’ve read my ultracode breakdown and want to go big with fan-outs, Pro is the wrong tier.
The same Claude Code everywhere you log in
Your Pro login works in the terminal CLI, the VS Code extension (and forks like Cursor), and JetBrains IDEs. Run /登入 with your Claude account and you’re on the subscription. My Claude vs Claude Code explainer covers how the chat app and the coding tool differ if that’s still fuzzy.
What Pro doesn’t include
- No monthly API credits. Max now comes with $100/month in Claude Platform API credits and Ultra with $200. Pro gets none. Don’t over-value those credits either: Anthropic’s support page says they 不能 be used inside Claude Code or the Claude apps.
- No priority access at high-traffic times. That’s Max and up.
- A thinner “wrap-up” allowance. If you hit the 5-hour limit while Claude Code is mid-response, it can finish the step it’s on (“Usage limit reached · wrapping up”). On Pro that grace applies once per weekly usage period. On Max it can apply every time you hit a 5-hour limit. Anthropic says it’s still rolling out, so you may not have it yet.
The Three Walls You Hit on Claude Code Pro

Anthropic doesn’t publish Pro’s limits in tokens or messages. The pricing page says limits depend on “the model you use, how long your chats are, and how large your files are”. So nobody can honestly tell you “Pro gets you X prompts”. What you 能 know is which wall shows up first, and the answer depends on how you work.
Wall 1: the 5-hour window
Pro and Max both have a rolling five-hour session limit. It’s the wall you’ll meet first if you work in long, focused blocks: a three-hour refactor on Opus, a big codebase read, a long debugging chase. When you hit it, you wait for the window to reset, turn on usage credits, or upgrade. Ultra is the only individual plan with no 5-hour limit, and only for verified accounts.
Wall 2: the weekly cap
Steady daily use hits this one. You might never max out a single 5-hour window but still run dry by Thursday. The weekly limit is the real measure of what Pro is worth to you, because it can’t be dodged by spreading work across the day.
Wall 3: parallel sessions
Every session draws on the same pool. Two terminals, a subagent exploring the repo, a workflow fanning out helpers: they all count. The way Claude Code feels best (several things running at once) is also the fastest way through a Pro allowance.
How to see the walls coming
跑步 /用法. On a subscription it shows plan usage bars, activity stats and, on recent versions, a breakdown of what’s eating your limit (skills, subagents, MCP servers) with suggested fixes. Press d 或者 w to switch between the last 24 hours and the last 7 days. I walked through that screen in detail in Claude Code usage, explained, so I won’t repeat it here. One warning from the docs: the Session cost figure on that screen is “intended for API users”. On Pro, it isn’t your bill. Keep that in mind, because it becomes useful in the break-even section below.

取得人工智慧策略手冊
The playbook behind my agent fleet: how I scope one job, what each run must report back, and the checks that catch the expensive mistakes. Free, and you can leave whenever you like.
Claude Code Pro Price vs Max vs Ultra vs an API Key: The Upgrade Triggers

Forget “which plan is best”. The Claude Code Pro price only looks small or large next to what you’d pay instead. Ask which symptom you have, then pay for the cheapest plan that fixes it:
| What keeps happening | Move to | Monthly cost | 為什麼 |
|---|---|---|---|
| You use Claude Code a few evenings a week and rarely see a limit | Stay on Pro (consider annual) | $20, or ~$17 on annual | You’re getting more than you’d pay at API rates. Don’t touch it. |
| You hit the 5-hour window once or twice a month | Pro + usage credits | $20 + whatever you top up | Paying per overflow beats paying $80 more every month for a rare spike. |
| You hit the 5-hour window most weeks, or you want Fable included | 最大限度 | $100 | 5x the 5-hour and weekly allowance, Fable up to half the weekly limit, priority access. |
| Long unattended runs, or the 5-hour reset breaks your day | Ultra | $200 | No 5-hour limit on verified accounts and 10x Pro weekly. The weekly cap is the only wall left. |
| Bursty use: nothing for weeks, then one heavy sprint | API key (Console, pay per token) | Whatever you use | No session caps and no monthly fee. You pay list rates, so measure first. |
| 2+ people need it with shared billing and SSO | Team | $20-$25 per standard seat | The price matches Pro, plus admin, SSO and no training on your content by default. |
注意一下什麼是… 不是 in that table: “upgrade because you might need it someday”. Max costs five times Pro. Ultra costs ten times. Both are monthly, so you can drop back down the day the crunch ends. That makes upgrading cheap to reverse, and it’s why I’d never pay for Max “just in case”.
If you’re a business trying to work out which tier your team needs before rolling Claude Code out, that’s the kind of call I help people make in a 戰略會議. We look at the real workload, not the pricing page.
For API pricing, the relevant list rates per million tokens (from Anthropic’s pricing docs, 22 November) are: Opus 5.5 $4 input / $20 output, Sonnet 5.5 $2 / $10, Haiku 5.5 $0.10 / $0.50 (prompts up to 100k tokens), and Fable 5.2 $8 / $40. Cache reads cost a fraction of that ($0.20 per million on Opus 5.5). Caching is why real Claude Code sessions cost less than the raw token counts suggest.
Why My Agent Fleet Isn’t on Pro (the Receipt)

The research brief for this post asked me to check one thing before using my fleet as an example: is it actually on Pro? It isn’t. Here’s what I found in the container on 22 November:

⚡ 取得人工智慧優勢
每週提供真正省時省錢的AI小技巧。沒有廢話,沒有誇大其詞——只有切實有效的方法。.
- It’s on a subscription, not an API key. The token the scheduler hands to Claude Code is a subscription OAuth token, and there’s no
人類學 API 金鑰set in the container. So every run draws on plan limits, not per-token billing. - The plan is the $200 tier. November’s Anthropic invoice reads “Claude Ultra”. It used to say “Max 20x” before Anthropic renamed the tier.
- The schedule runs all day. This container has 12 daily jobs, from blog-writer at 00:07 to the watchdog at 23:37, plus 10 weekly ones. From 07:07 until midnight a new job starts at least every two hours. Over 30 days (21 October to 20 November) that came to 415 runs, which I broke down in my AI implementation consultant post. 。 這 克勞德代碼作弊表 has the earlier month.
Put that schedule on Pro and the 5-hour window becomes a problem nobody is awake for. The docs are blunt about what happens to unattended work at a limit. A dynamic workflow pauses and waits for the reset only in an interactive terminal session or a local Code-tab session. Under 克勞德-p, the Agent SDK or a scheduled task, the agent that hits the limit 失敗 instead. A cron job has nobody at the keyboard to click “continue”. It just stops, and you find out in the morning.
That’s the real reason I pay for Ultra, and it isn’t raw volume. Ultra removes the 5-hour wall, and unattended jobs can’t wait out a wall on their own. If your use is one person at a keyboard, that argument doesn’t apply to you, and Pro may be plenty.
One honest gap: I can’t tell you what each fleet run would cost at API rates. Claude Code isn’t keeping session transcripts for the main project in this container, so the full token counts aren’t on disk. I did find one small transcript (more on it next), and fixing the logging is on my list. It isn’t done.
Break-Even Math You Can Run on Your Own Account

Is the Claude Code Pro price worth it for you? Other pricing guides already do the per-day math: $20 / 30 is about $0.67 a day, so if your API-rate spend stays under that, a key is cheaper. That’s correct, but it skips the hard part, which is knowing your API-rate spend without first paying API rates. Here’s how I’d measure it, starting with the one real call I could price.
A measured call, priced at list rates
On 21 November at 20:05, my social-poster job asked Claude Code (Opus 5.5) to write a 7-beat video script. That nested call left a transcript, and this is its usage block:
| Token type | Tokens | Opus 5.5 list rate | 成本 |
|---|---|---|---|
| Output (1,802 of them thinking) | 2,272 | $20 / M | $0.045 |
| Cache write (1-hour) | 8,902 | $8 / M | $0.071 |
| Cache read | 10,480 | $0.20 / M | $0.002 |
| Uncached input | 2 | $4 / M | ~$0 |
| 全部的 | ~$0.12 |
Twelve cents. On an API key, $20 (the Claude Code Pro price) would pay for roughly 168 calls like that. Two lessons hide in the table. First, the cache 寫 cost more than the output. Short one-shot calls never get to reuse what they cached, so they pay full price to set it up. Second, thinking tokens bill as output: 1,802 of the 2,272 output tokens were thinking. If you run high effort settings, your output bill is mostly reasoning you never read.
The test for your own account
- Use Claude Code normally for a week on whatever plan you’re on.
- Read the Session cost in
/用法at the end of each session. Claude Code computes it locally from token counts at list price. It isn’t your subscription bill, which makes it exactly the “what would this cost on an API key” number you need. Note that/清除resets it (v2.1.211+), so read it before you clear. - Add up the week and multiply by 4.3. That’s your monthly API-rate equivalent.
- Compare. Under $20: an API key is cheaper, as long as you’re disciplined. $20-$100 with no limit hits: Pro is a bargain. Over $100 and hitting walls weekly: price Max against your number.
For calibration, Anthropic’s own cost docs say enterprise Claude Code users on API billing average about $13 per developer per active day and $150-$250 a month. If you look anything like that, Pro’s $20 covers less than two days of API-rate usage. That’s why it feels like a steal right up until the first wall.
Ways People Overpay on Top of the Claude Code Pro Price

1. A stray ANTHROPIC_API_KEY
This is the expensive one. Anthropic’s support article warns that if 人類學 API 金鑰 is set in your environment, Claude Code uses the key 而不是 your subscription, so you pay API rates while also paying $20 a month. It’s easy to do by accident: a key exported in .zshrc from an old SDK project, or a .env file your shell loads. Anthropic points to /地位 for checking what you’re running on. If you only want plan usage, Anthropic’s advice is to log out, log back in with only your Claude subscription, and skip adding Console credentials.
2. “Group buy” resellers
When this keyword was researched on 16 November, one of the pages ranking for “claude code pricing” was selling “group buy” access, meaning shared accounts at a discount. Sharing one subscription between strangers is the opposite of what the plan is built for. You’re trusting a stranger with your code and your session. At $17 a month on annual billing, the official price is already low. Pay it.
3. Annual Pro you’ll outgrow by March
$200 up front saves $40 a year, but only while you stay on Pro. If you’re already hitting the 5-hour window, you’re one busy month away from Max, and Max is monthly only, so the annual discount doesn’t come with you. My rule: run monthly for two months, check /用法, then commit.
4. Max “just in case”
Max is five times the price of Pro. If you hit a wall once a month, usage credits cover the overflow for far less. Upgrade when the symptom shows up every week, not before. The Claude Code Pro price plus a few dollars of credits is usually still the cheapest line on the bill.
5. A heavy CLAUDE.md on every request
Your memory file rides along with every request. The /用法 recommendations will literally say things like “Trim CLAUDE.md · 49k tokens per request”. On Pro, that bloat isn’t a line on your bill. It’s a wall you reach sooner. My Claude Code CLI operator guide covers keeping that file lean.
Claude Code Pro Price FAQ

Claude Code包含在Claude Pro中嗎?
Yes. Claude Code is listed as a Pro feature on the pricing page, and one subscription covers both claude.ai and Claude Code, including the IDE extensions. The usage limits are shared between them.
How much is Claude Code Pro per year?
$200 billed up front, which works out to about $17 a month, against $240 for twelve monthly payments. Max and Ultra don’t offer annual billing.
Does Claude Code Pro include Opus?
Yes. Opus, Sonnet and Haiku are all included on Pro. Fable isn’t included in the plan’s limits on Pro. You pay for it with usage credits.
What happens when I hit the Pro limit mid-task?
Claude Code may finish the step it’s on (the wrap-up allowance, once per weekly period on Pro, still rolling out). After that you can wait for the reset, enable usage credits to keep going, or switch to a Console account and pay API rates for a sprint.
Is a Team standard seat better than Pro?
For one person, usually not. The standard seat is the same $20 on annual billing (or $25 monthly) with “more usage than Pro”. What you’re really paying for is admin, SSO, central billing and no training on your content by default. Once two or more people need it, Team is the better deal.
Can I use the Max or Ultra API credits for Claude Code?
No. Anthropic says the monthly API credits included with Max and Ultra can’t be used in Claude Code or the Claude apps. They’re for building on the Claude Platform.
Final Thoughts: Pay for the Wall You Actually Hit
The Claude Code Pro price is the easy part: $20 a month, or $17 if you commit for a year. What you’re really choosing is a set of walls. Pro gives you a 5-hour window, a weekly cap, workflows sized for one person, and Fable on credits. For someone coding a few hours a day at a keyboard, that’s one of the best deals in software. Your own /用法 numbers will prove it inside a week.
Move up only when a specific wall keeps hitting you. Weekly 5-hour hits mean Max. Unattended or overnight work means Ultra, because a cron job can’t wait out a reset. Bursty, occasional use means an API key. And whatever you pick, check that a stray 人類學 API 金鑰 isn’t quietly billing you on top of it.
Start with one week of measuring, not a guess. Then pay for the wall you actually hit.

取得人工智慧策略手冊
每週:我交給經紀人的一個真實項目,費用是多少,以及哪些環節出了問題。絕無廢話,隨時可以取消訂閱。.

📥 免費:《人工智慧劇本》
我用來經營一人代理公司的所有工具和工作流程。 25 年的行銷經驗濃縮成一份實用指南。免費贈送。.
