# Claude Code Pricing: How to Pay Cost-Effectively and Choose a Model > Detailed analysis: How to pay for Claude Code cost-effectively in 2026: plans and models. Source: https://ailoot.ai/blog/en/claude-code-plans-cost/ Canonical: https://ailoot.ai/blog/en/claude-code-plans-cost/ Language: en Publisher: Ailoot (https://ailoot.ai/) Translation: guide-132 Revision: sha256-422307f65d9b2116935f4fde36e0ec4b96595aac4da9311d73406567750ec7e3 “Pay for Claude Code” is a search for access to a model. Behind it is a simple desire: pay less for the tool without losing work quality. Search results answer this poorly: services acting as middlemen invite you to pay through them, while almost no one explains what the money goes toward or where the overpayment is hidden. This analysis covers everything in order and with figures. What you are buying, which payment arrangement costs less for your work pattern, five decisions that cut the bill, and how to check the payment terms. Before publication, the prices in the source material were checked against the official Anthropic, DeepSeek, and Z.ai pages. For additional instructions, see the [Ailoot documentation](https://ailoot.ai/docs/).
Attention This material is for informational purposes and is not legal advice.
## What are you buying when you pay for Claude Code?
Key takeaway You cannot pay for Claude Code as a standalone program: it is free, and you pay for access to the model. Access is sold in two ways—through a Claude subscription with a fixed monthly fee, or through a prepaid balance in Claude Console, where you are charged for each request. The free plan does not provide access to Claude Code.
Claude Code itself is free. It is installed with one command from Anthropic's official site: ```bash curl -fsSL https://claude.ai/install.sh | bash ``` On Windows, run this command in PowerShell or inside WSL—the build is available for both. After installation, the program asks you to sign in to an account. The documentation says:

Claude Code requires a Pro, Max, Team, Enterprise, or Console account. The free Claude.ai plan does not provide access to Claude Code.

— Claude Code documentation, https://code.claude.com/docs/en/setup
That is why half the questions in search are phrased incorrectly. You cannot “top up Claude Code”—the program has no balance of its own. You can top up your Claude Console account if you work through the API. “Pay for a Claude Code subscription” is also imprecise: there is one subscription for everything, and it gives you access to the web version, the app, and terminal use all at once. The two payment models work differently: | What you pay for | How you are charged | Who it suits | |----|----|----| | Claude subscription (Pro, Max, Team) | Fixed monthly fee, with time-based limits | You work hands-on every day and want a predictable bill | | Claude Console balance | You add funds in advance and are charged for actual requests | You work in bursts: one active month, then two months without using it | If you are searching for “pay for Claude Code” using a transliteration of its name, it means the same thing: Claude Code. Anthropic has no separate product by that name, in any spelling. ## How much does Claude Code cost in 2026?
Key takeaway Pro costs 20 dollars per month or 200 dollars upfront for a year, which works out to 17 dollars per month. Max is available only month to month: for 100 dollars, its limits are five times higher than Pro's; for 200 dollars, they are twenty times higher. Team costs 25 dollars per seat. Claude Code is included in all paid plans, including basic Pro.
The prices in the source calculation were checked against Anthropic's official pricing page and help center on August 22, 2026: | Plan | Monthly price | What it includes | |----|----|----| | Free | \$0 | Claude Code not included | | Pro (monthly) | \$20 | Claude Code, Cowork, unlimited projects | | Pro (annual, paid upfront) | \$17 | Same features; \$200 charged at once | | Max 5x | \$100 | Limits 5 times higher than Pro | | Max 20x | \$200 | Limits 20 times higher than Pro | | Team (monthly) | \$25 per seat | Shared bill, single sign-on, 2 to 150 seats | | Team Premium | \$125 per seat | Limits 5 times higher than a regular seat | The help center has a detail about Max that many people miss: it is available only month to month, with no annual payment option. You cannot ask for a discount either—Anthropic explicitly says its support team does not offer discounted prices or coupons. Usage within the plan is shared: limits are split across the web version, app, and terminal. Open a long Claude Code session and you will hit the cap earlier than usual in your browser that same day. I explain exactly what each plan includes and where to check your remaining usage in the article about [Claude Pro and Max limits](https://ailoot.ai/blog/en/claude-pro-max-usage-limits/). Many people choose Team for a shared bill, which is more convenient for accounting. The plan is designed for groups of 2 to 150 seats. A monthly seat costs 5 dollars more than Pro; with annual billing, the gap shrinks to three dollars: 20 versus 17. ## Which costs less: a subscription or pay-as-you-go?
Key takeaway A subscription is a better deal when you work with the higher-tier models every day: at API prices, the same 20 dollars gets you less than three million tokens on Opus. Pay-as-you-go wins for an intermittent work pattern and simple tasks where a lower-tier model is enough. Console funds expire one year after purchase and are nonrefundable.
API prices are listed per million tokens. Prices current as of August 22, 2026: | Model | Input per 1M | Output per 1M | |-------------------|------------|-------------| | Claude Opus 4.8 | \$5 | \$25 | | Claude Sonnet 5 | \$2 | \$10 | | Claude Sonnet 4.5 | \$3 | \$15 | | Claude Fable 5 | \$10 | \$50 | The Fable 5 row is now mostly historical: Anthropic discontinued the model three days after its release. I covered that story in my note about [the discontinuation of Fable 5 and Mythos 5](https://ailoot.ai/blog/en/claude-fable-mythos-model-switch/). The working pair today is Opus and Sonnet. Here is a calculation worth doing before you pay. In an agent's work, input greatly outweighs output: the program reads files, history, and instructions, then writes back relatively little. I am using a nine-to-one ratio and calculating for a full million tokens. With Sonnet 5, the cost is 0.9 × 2 + 0.1 × 10 = 2.80 dollars per million. A 20-dollar Pro subscription covers about seven million tokens. With Opus 4.8, the same million costs 0.9 × 5 + 0.1 × 25 = 7 dollars, so 20 dollars covers less than three million. The actual bill can be lower when repeated input qualifies for cache reuse, but cache reads, initial writes, output, and repeated turns all count. A longer session does not guarantee savings: it also accumulates history. Confirm reuse from usage telemetry and compare total processed tokens for the completed task. The article about [Claude Code token usage](https://ailoot.ai/blog/en/where-claude-code-tokens-go/) breaks down where the tokens go. Funds in your Console balance do not last forever:

Credits expire one year from the date of purchase, and this period cannot be extended. All credit purchases are nonrefundable.

— Claude Help Center, https://support.claude.com/en/articles/8977456-how-do-i-pay-for-my-claude-api-usage
Anthropic calls the prepaid balance “credits,” though they have nothing to do with borrowing. Failed requests are not charged: only successful calls and completed tasks are billed. The model bill is not the only line item in your expenses; the full analysis is in [why vibe coding costs more](https://ailoot.ai/blog/en/vibe-coding-cost-time/). ## How to get the best deal: five decisions in order
Key takeaway Five choices can make Claude Code cheaper: choose an annual Pro subscription instead of monthly billing, use a lower-tier model for routine work, keep related work together while bounding history and checking cache reuse, move to Max only when you hit Pro's cap, and subscribe through your own account. A Team seat makes sense if you have a real group of at least two people.
Here they are, from simplest to most cost-effective. **1. Pay for Pro annually.** Pay 200 dollars upfront instead of 240 across twelve monthly charges. That saves 40 dollars a year. The obvious drawback is that the money is gone upfront for a full year, and you cannot get it back; more on refunds below. **2. Use a lower-tier model for routine work.** At API prices, Sonnet 5 costs two and a half times less than Opus 4.8 for both input and output. It can handle edits based on examples, tests, renaming, and log analysis. For where to draw the line between models, see [how to choose Sonnet or Opus](https://ailoot.ai/blog/en/claude-code-model-effort-sonnet-opus/). **3. Reuse stable project context without growing history indefinitely.** Continue a session while its recent discussion is relevant to the same task. For a new task, save a short handoff and load only the needed files in a fresh session. A new session does not by itself prove a cache miss, and staying in one does not prove a hit: reuse depends on matching input and provider cache conditions. Measure cached input and the total input/output across all turns before choosing. If Claude keeps starting from scratch to understand your project, that can be fixed; I have a separate article on [why Claude Code rereads the whole project](https://ailoot.ai/blog/en/claude-code-rereads-project/). **4. Move to Max only after Pro.** It makes sense when you regularly hit Pro's cap several days a week. Once a month is not enough: the plans differ by 80 dollars a month. The source author paid for Max 20x because they worked in the terminal for several hours every day. Before that, Pro had been enough for six months. **5. Subscribe through your own account.** This is not about a discount; it is about the cost of a mistake. Someone else's account and paying through a middleman may look cheaper than the official price, but the Anthropic agreement is with the account holder. If the account is closed, you lose both access and the money for the unused period. More detail is in the section on what not to do. A Team seat costs 25 dollars, five more than Pro. It makes sense only for a group: the plan starts at two seats and provides a shared bill with single sign-on, so for one person it is simply an extra expense. The list deliberately leaves out tactics that promise the same access below the official price. Such offers always involve either someone else's account or a middleman, and I explain below how that ends. Your Claude Code bill depends only half on the plan. The other half is how much extra the program rereads with each request and how many times it redoes the same work. This is context engineering. The plan stays the same, while the amount Claude rereads for each request goes down. ## Why is a card from an unsupported country declined? In Anthropic's help center, geography is listed as the first reason for a decline:

Check the list of supported billing countries and make sure the billing address for your payment method and the country where it was issued are eligible for payment processing.

— Claude Help: why was my card declined, https://support.claude.com/en/articles/9402418-why-was-my-card-declined
The check looks at both the country where the card was issued and the billing address. Changing browsers, clearing your history, or trying a third time is pointless; the same check will run again. The second item on the same list is a mismatch between the billing address and the address on file with your bank. Anthropic warns that even a small detail can cause problems: a missing diacritic or a typo in the street name. Next come an unverified payment (the bank sends a code by text, but you do not enter it), an unsupported payment method, insufficient funds, and temporary network issues. That leads to the main mistake: entering a billing address that is not yours. The agreement requires accurate billing information. The mismatch is uncovered during the first bank check, and the account is closed along with the paid period. In that case, the money is not refunded. ## What payment methods does Anthropic officially accept?
Key takeaway For web subscriptions, Anthropic accepts credit and debit cards only. Payment methods in the iOS and Android apps are set by the app store. Team accepts cards, including prepaid cards, but not bank transfers. Enterprise customers pay by bank transfer. Third-party payment services acting as middlemen are not accepted anywhere.
Anthropic has a separate help article for each account type. Together, they add up to this: | Account type | What is accepted | |----|----| | Pro and Max through the website | Credit and debit cards only | | Pro and Max through the app | Methods available in the App Store or Google Play | | Team | Credit, debit, and prepaid cards; bank transfers are not accepted | | Enterprise (self-serve) | Card or bank transfer within the United States (ACH) | | Enterprise (through sales) | Bank transfer; invoices of 50,000 dollars or more must be paid by transfer | | API in Console | Card; add funds to the balance in advance | Third-party payment services acting as middlemen are not accepted on any plan. This appears as a separate item on the list of reasons for a decline. Prepaid cards are listed—but only for Team. The billing-country requirement applies to them in exactly the same way. Payment through the app works differently: Apple or Google charges you according to its own rules. If you are asked to verify your identity at some point, there is a separate article about [what happens with account verification](https://ailoot.ai/blog/en/claude-identity-verification-persona/). ## Can you pay through Amazon, Google, or Microsoft?
Key takeaway Claude Code officially works through accounts other than Anthropic, too. The documentation names three cloud platforms: Amazon Bedrock, Google Cloud, and Microsoft Foundry. The model is the same, the cloud provider sends the bill, and each has its own payer requirements.
The sign-in section of the documentation lists them in a row:

You can use Claude Code with a third-party API provider, such as Amazon Bedrock, Google Cloud Agent Platform, or Microsoft Foundry.

— Claude Code documentation, section on signing in, https://code.claude.com/docs/en/setup
Here is how billing works in this case: 1. ### Register with a provider Create an account with the cloud provider and complete its verification. 2. ### Get access to models Enable Claude models in the provider's console. 3. ### Switch Claude Code Tell Claude Code which cloud platform to use. Anthropic's documentation lists the exact variables for each platform. 4. ### Get billed The cost is added to the provider's overall bill along with its other services. AWS and Microsoft Foundry calculate Claude usage in their own units: one hundred units equal one dollar at standard Anthropic prices. You pay for actual usage, so you do not need to add funds in advance. Google Cloud has its own prices: tying your setup to a specific region makes it 10% more expensive than ordinary access. The cloud removes your dependence on an Anthropic subscription, but you still need a card or an invoice. Amazon, Google, and Microsoft also check your country and require a bill they are willing to process. For a legal entity with an overseas bank account, this is the easiest way to account for expenses. For an individual without a legal entity, this route is usually harder. ## How do you switch Claude Code to another model if payment fails?
Key takeaway If you cannot pay Anthropic, you can keep the program and change who responds to your requests. Claude Code reads two environment variables: the server address and the key. Z.ai and DeepSeek have set up APIs that Claude Code understands and published setup instructions.
The first variable sets the address requests are sent to; the second passes the key. Here is Z.ai's official instruction: ```bash export ANTHROPIC_AUTH_TOKEN=your_zai_key export ANTHROPIC_BASE_URL=https://api.z.ai/api/anthropic ``` DeepSeek has its own address, and the variable for the key is named differently: ```bash export ANTHROPIC_API_KEY=your_deepseek_key export ANTHROPIC_BASE_URL=https://api.deepseek.com/anthropic ``` Paste both lines directly into the terminal before starting claude. They work only while the window is open: close the terminal and the settings disappear. To avoid entering them every time, add the same lines to the end of ~/.zshrc (on Mac) or ~/.bashrc (on Linux), then restart the terminal. DeepSeek also remaps model names on the fly: requests to Opus go to deepseek-v4-pro, while requests to Sonnet and Haiku go to deepseek-v4-flash, so you do not need to rewrite the settings for each model. What changes and what stays the same: | Stays the same | Changes | |----|----| | Program, commands, familiar keyboard shortcuts | Response quality on complex architecture | | Project rules file and folder structure | Context window length | | Skills, subagents, connected tools | Speed and behavior on long tasks | | Your projects on disk | Who receives your files and requests | When you change the server address, you also change who can see the project contents: files the agent reads go to the new provider. The source author did not switch their work project to an external server, but did switch personal projects because they had no sensitive data. For exactly what is sent and how to limit it, see the article about [data security in Claude Code](https://ailoot.ai/blog/en/claude-code-data-access-safety/). If you would rather use a complete replacement, see these detailed analyses: [Qwen Code as a Claude Code competitor](https://ailoot.ai/blog/en/qwen-code-token-plan-setup/), [ZCode on GLM-5.2](https://ailoot.ai/blog/en/zcode-claude-code-alternative/), and the overview of [Claude Code alternatives](https://ailoot.ai/blog/en/claude-code-alternatives/). ## How much do Chinese models in the same class cost?
Key takeaway The GLM Coding Plan subscription starts at about 18 dollars a month—comparable to Claude Pro. Savings begin with the API: a million tokens on deepseek-v4-pro costs 0.79 to 1.58 dollars, depending on the time of day, compared with seven dollars on Opus 4.8. That is a difference of four to nine times.
The ratio is the same: nine to one. Since summer 2026, DeepSeek has charged different prices at different times of day; rates double during peak hours. Peak hours in UTC are 01:00 to 04:00 and 06:00 to 10:00. Starting August 23, weekends are entirely billed at the off-peak rate. Prices per million tokens: | Model | Input per 1M | Output per 1M | Million tokens at 9:1 | |---------------------------|------------|-------------|-------------------------| | Claude Opus 4.8 | \$5 | \$25 | \$7.00 | | Claude Sonnet 5 | \$2 | \$10 | \$2.80 | | deepseek-v4-pro (off-peak) | \$0.66 | \$1.98 | \$0.79 | | deepseek-v4-pro (peak) | \$1.32 | \$3.96 | \$1.58 | | deepseek-v4-flash (off-peak) | \$0.22 | \$0.66 | \$0.26 | | deepseek-v4-flash (peak) | \$0.44 | \$1.32 | \$0.53 | DeepSeek’s table also lists a lower cached-input rate: from 0.022 dollars per million on the pro model. Apply it only to input reported as a cache hit. Session length alone does not establish a larger saving over Anthropic; compare measured cache use and total task cost under the applicable tariffs. Subscriptions work differently. Z.ai's documentation describes the GLM Coding Plan as a package for terminal work: you pay a fixed amount and get a monthly request allowance. At publication, the entry tier was about 18 dollars a month and included GLM-5.2, GLM-5-Turbo, and GLM-4.7. You can use it from the same Claude Code. These figures do not mean everyone should switch. Opus is still stronger at architecture analysis and long chains of reasoning, and code rewritten three times eats up all the savings. The difference shows up in routine work: edits based on a finished example, tests, renaming, and log analysis. The author compared these models in real work in separate articles: DeepSeek versus Claude Code and [GLM-5.2 versus Claude](https://ailoot.ai/blog/en/glm-5-2-vs-claude-code/). For a general picture of coding tools, see the review of [what to code with in 2026](https://ailoot.ai/blog/en/ai-coding-tools/) and the collection of [free models for coding](https://ailoot.ai/blog/en/free-coding-models/). ## What should you avoid when paying?
Key takeaway Four things reliably end in lost money: buying someone else's account, paying through middlemen, retrying a declined payment, and choosing an expensive plan on day one. The first takes away your rights to the account, the second offers no protection, the third repeats the decline, and the fourth makes a mistake ten times more expensive.
**Someone else's account.** The Anthropic agreement is with the person who opened the account. Buying access does not make you the owner: the password can be changed at any time, and there may be no one to help you recover access. If the account is closed, the conversation history and settings go with it. That is why you should subscribe through your own account, even if it is harder: your account means your money, your history, and the ability to contact support and be heard. **Paying through middlemen.** No middleman appears among the payment methods Anthropic itself lists. You hand over either your card details or account access, and there is no protection either way. In addition, this violates the terms of service. The account may then be closed, with no refund for the remaining period. There will be no one to complain to and nowhere to complain. **Retrying after a declined payment.** If the country is the reason for the decline, retrying changes nothing. After several failed attempts, however, your bank may temporarily restrict card transactions. The first thing to check is whether the billing address matches the one on file with your bank. **Choosing an expensive plan on day one.** If you discover after paying that the terms are not right, it is harder to get 200 dollars back than 20, so start with Pro and upgrade when you hit the cap. A separate note on money already spent: Console prepaid balances are nonrefundable—that is stated plainly in the help center. The only safeguard is not to put more in the account than you expect to spend in a year. ## You paid, but the limit ran out in two hours. What now?
Key takeaway This can happen on any plan. Your usage depends on how much the program rereads for each request. With careful use, Pro usually lasts a full workday, as long as you do not make Claude figure out the project from scratch every time. When the limit runs out, you can wait for it to reset, upgrade, or enable pay-as-you-go through Console.
Each request sends the model your message along with all the accumulated context: files it has read, session history, and project instructions. The longer the session, the heavier each subsequent request gets. Open a dozen files and keep chatting in the same window, and you pay for them again and again. That is how a subscription gets used up in one day, and it is the most common complaint from new users. What helps: - Keep a file with project rules so Claude does not have to work out the structure again. See the article about [a project's Second Brain](https://ailoot.ai/blog/en/claude-code-second-brain/) for how to put one together. - Start a fresh, bounded session when the task changes or old history stops helping. Save the current result and next action first. Continue the existing session when its recent context is still useful; compare total usage rather than assuming either option always wins. - Use a simpler model for routine work and save the higher-tier one for architecture. - Track usage in numbers. See the collection of token tracking tools for ways to measure it, and the analysis of [how not to burn through Claude Code limits](https://ailoot.ai/blog/en/claude-code-token-limits-saving/) for eight savings rules. When the limit does run out, the help center suggests waiting for the reset, upgrading, or enabling additional credits. Another option is to switch to pay-as-you-go through a Console account. It is billed at standard API prices, which are higher than subscription prices, and requires explicit confirmation: you cannot switch to it accidentally. ## What else do people ask about paying for Claude Code?
Key takeaway Below are brief answers to the questions people most often search alongside the main topic: how the name is transliterated, refunds, canceling a subscription, and tax on the receipt. Each answer stands on its own; you do not have to read them in order.
### Frequently asked questions Can you get a refund for a Claude subscription? Under Anthropic's terms of service, all payments are nonrefundable unless a refund is required by law. If you cancel your subscription, access continues until the end of the paid period, but you will not get the payment back. The agreement has one exception: if Anthropic changes the terms and you terminate your subscription before the changes take effect, you will receive a partial refund for the remaining period. The practical conclusion is simple: do not pay upfront for more than you are willing to spend without a refund. What happens to Claude Code if you cancel your subscription? Access remains until the end of the period you have already paid for. After that, Claude Code will no longer let you use the model: the program is free, but it does not work without a paid account. Your account and conversation history remain, and you can return at any time by subscribing again. Why is it better to subscribe to Claude through your own account? Because the Anthropic agreement is with the account holder. On someone else's account, your money, history, and settings belong to another person: they can change the password and you will lose everything, including the paid period. With your own account, you have support, payment history, and a way to recover access. Why is the amount on the bill higher than the price on the plans page? Tax is added according to the rules of the country on your billing address. Anthropic calculates it automatically based on the address you provided when paying and shows it as a separate line on the receipt.
## How to get started
Key takeaway The steps are simple: decide what to pay for—a subscription or a Console balance—check your billing address against Anthropic's country list, and start with the lowest-priced plan. If payment fails, switch Claude Code to another model provider and keep working while you sort out the billing issue.
1. ### Calculate your workload How many hours a week will you actually spend in the terminal? Up to five, and Pro should be enough; beyond that, consider Max 5x. 2. ### Check your billing address Check Anthropic's list of supported countries before entering your card details. 3. ### Start with the lowest-priced plan Upgrade when you hit the cap. Annual Pro saves 40 dollars, but charges you upfront. 4. ### Do not retry a declined payment If the country is the reason, no number of attempts will change it. After several failed charges, your bank may temporarily restrict card transactions. 5. ### Prepare your project before the first request While the payment issue is unresolved, switch Claude Code to another model provider and keep working. A paid plan by itself does not solve anything: the same Pro plan lets one person's work project run smoothly, while another burns through it in two hours redoing work. The difference is what is in the project before the first request.
### Sources - [Claude Pricing](https://claude.com/pricing) - official plans page, checked 22.08.2026 - [What is the Max plan](https://support.claude.com/en/articles/11049741) - Max 5x and Max 20x pricing, monthly only - [Claude Code: Advanced setup](https://code.claude.com/docs/en/setup) - account requirements and cloud providers - [Supported countries and regions](https://support.claude.com/en/articles/8461763-which-countries-and-regions-can-i-use-claude-ai-from) - list of supported countries - [Why was my card declined](https://support.claude.com/en/articles/9402418-why-was-my-card-declined) - reasons for payment declines - [How do I pay for my Claude API usage](https://support.claude.com/en/articles/8977456-how-do-i-pay-for-my-claude-api-usage) - prepaid balance and credit expiration - [Anthropic Consumer Terms of Service](https://www.anthropic.com/legal/consumer-terms) - refund and subscription cancellation terms - [Anthropic Pricing](https://www.anthropic.com/pricing) - model prices per million tokens, checked 22.08.2026 - [DeepSeek: Models & Pricing](https://api-docs.deepseek.com/quick_start/pricing) - peak and off-peak prices, checked 22.08.2026 - [DeepSeek: Anthropic API](https://api-docs.deepseek.com/guides/anthropic_api) - compatibility with Claude Code - [Z.ai: connect Claude Code](https://docs.z.ai/scenario-example/develop-tools/claude) - official instructions - [Z.ai: GLM Coding Plan subscription](https://z.ai/subscribe) - subscription plans
When you are ready, get to know [Ailoot](https://ailoot.ai/). ## Questions about How to Pay for Claude Code Cost-Effectively in 2026: Plans and Models ### Should I keep one session open all day to save on cached input? Continue while the recent context helps the same task. When the task changes or history grows noisy, save a compact handoff and start fresh with the required files. Neither session length nor a new session proves a cache hit or miss; check usage telemetry. ### Which usage figures should I compare before changing my workflow? Compare total input, cached input, output, and repeated turns for the same completed task. Include any cache-write cost. A high cache-hit share can still accompany a large bill if every turn rereads a growing history. ## Ailoot · The company and its project portfolio Ailoot combines established operating experience, a large client base, stable service and ongoing KPI control. The company and its project portfolio: millions of customers; monthly sales in the billions. Ailoot brings that operating experience to API access for AI features in customer products, agents, applications and automations.