Claude Pricing Explained: Every Plan and Every API Rate (2026)

How to read the numbers on this page

Anthropic sells Claude in three structurally different ways, and most confusing Claude AI pricing comparisons come from mixing them up. Anthropic Claude pricing is also published across three separate places, so figures drift between them. There is no single cheapest option: the answer depends on whether your usage is steady or spiky, whether a person or a program is asking, and whether anyone needs admin controls.

• Flat seat subscriptions: Free, Pro, Max 5x, Max 20x and Team. A fixed fee per person, with usage bounded by rolling windows rather than metered. Predictable bill, hard ceiling.
• Seat plus metered usage: Enterprise. The Claude pricing page lists self-serve Enterprise at USD 20 per seat, and the Anthropic Help Center states that usage is not included and is billed at standard API rates on top.
• Pure metering: the Claude API. No seat fee and no usage window, priced per million input and output tokens by model. Unbounded bill, no ceiling.
• Modifiers stacked on the API rate: prompt caching, the Batch API, data residency routing, fast mode and the tokenizer a model version uses. On real workloads these move the bill more than the choice of model does.
• What we checked and when: plan names, monthly and annual prices, seat minimums and the per-million-token input and output rate for every model on the current card, read on August 28, 2026.

Treat all of it as a snapshot. Anthropic revises this pricing every few weeks, retires model versions, and has cancelled a scheduled rate increase mid-year. Any Claude pricing figure you read anywhere, including here, is worth re-checking on the vendor page before you enter a card.

Claude pricing plans at a glance

Six paid surfaces exist alongside the free one, and the ladder is not linear: Pro to Max 20x is ten times the money for twenty times the session usage, while Team to Enterprise swaps a capped subscription for an open meter. Here are the Claude plans and pricing in one place, read on August 28, 2026.

• Free: USD 0. Claude on web, desktop, iOS and Android, with web search, memory, file creation, code execution, connectors and extended thinking, at limited usage.
• Pro: USD 20 a month, or USD 200 billed up front for a year. Adds Claude Code, Claude Cowork, Claude Design, Claude Science, unlimited projects and Research.
• Max 5x: USD 100 a month, billed monthly. Five times the Pro usage per five hour session, higher output limits and early access to new features.
• Max 20x: USD 200 a month, billed monthly. Twenty times the Pro usage per five hour session, plus priority access during high traffic.
• Team standard seat: USD 25 per member a month billed monthly, or USD 20 billed annually. Minimum two members, maximum 150 seats.
• Team premium seat: USD 125 per member a month billed monthly, or USD 100 billed annually. Same controls, significantly more usage per seat.
• Enterprise: USD 20 a seat self-serve plus usage at standard API rates, from 20 seats. Sales-assisted Enterprise starts at 50 seats and is quote-only.
• Claude API: no seat fee. USD 1 to USD 10 per million input tokens depending on model, and exactly five times that on output.

One pattern is worth carrying into every comparison below: every model on the current Claude API card prices output at exactly five times input. That ratio is why output-heavy work blows up API budgets and input-heavy work does not, and it is the first thing we teach about cost estimation in the free courses at Agentic School.

Claude Free and Claude Pro pricing

Claude Free costs USD 0 and Claude Pro pricing is USD 20 a month, and for most first-time buyers Free is enough to find out whether Pro is needed. The Anthropic Help Center describes Free as limited and suited to occasional use, and describes Pro as at least five times the usage per session compared with the free service.

Claude Free.

• Price: USD 0.
• Billing: none, no card required.
• Included: Claude on web, desktop, iOS and Android, code generation, web search, memory, file creation, code execution, desktop extensions, Slack and Google Workspace integration, connectors and extended thinking.
• Usage: limited, described by the Anthropic Help Center as suited to occasional use. No published message or token cap.
• Watch out for: Claude Code, Cowork, Design and Science are not on Free. If those are why you came, Free will not answer your question.

Claude Pro.

• Price: USD 20 a month, or USD 200 billed up front annually, which the pricing page displays as USD 17 a month and works out to about USD 16.70.
• Billing: monthly or annual.
• Included: everything in Free, plus Claude Code, Claude Cowork, Claude Design, Claude Science, unlimited projects, Research, multiple Claude models and Microsoft 365 integration.
• Usage: at least five times the free service per session, with the session limit resetting every five hours and a separate weekly limit across all models.
• Watch out for: Pro does not include Claude API access, which is a separate account and a separate bill. Extra usage can be bought on top, but no rate for it is published.

Claude Max pricing: the 5x and 20x tiers

Claude Max pricing is USD 100 a month for Max 5x and USD 200 a month for Max 20x, both billed monthly, per the Anthropic Help Center on August 28, 2026. The multiplier in the name is the usage multiple over Pro per session, not a performance claim. You get the same models, you simply hit the wall later.

Claude Max 5x.

• Price: USD 100 a month.
• Billing: monthly. No annual Max price is published, so do not plan around a yearly discount.
• Included: everything in Pro, plus higher output limits, early access to new features and priority access during high traffic.
• Usage: five times the Pro allowance per five hour session, plus the weekly limit across all models, which resets at a fixed time assigned to your account.
• Watch out for: the weekly limit is what surprises people. Clearing a five hour session limit does not mean the week is unlimited.

Claude Max 20x.

• Price: USD 200 a month.
• Billing: monthly, same as Max 5x.
• Included: an identical feature set to Max 5x. The only published difference is the usage tier.
• Usage: twenty times the Pro allowance per five hour session, with the same weekly limit layered on top.
• Watch out for: USD 200 a month is USD 2,400 a year for capacity, not capability. If you are not hitting Max 5x limits weekly, this is the easiest line item to over-buy in the whole lineup.

Claude Team pricing and Claude Enterprise pricing

Claude Team pricing starts at USD 20 per member a month billed annually, and Claude Enterprise pricing starts at USD 20 a seat plus metered usage, which means the cheaper-looking headline is the one that can cost more. Team is a capped subscription. Enterprise is a seat licence with an open meter attached.

Claude Team.

• Price: standard seats at USD 25 per member a month monthly or USD 20 annually. Premium seats at USD 125 monthly or USD 100 annually.
• Billing: monthly or annual, minimum two members, maximum 150 seats. Above 150 the Anthropic Help Center directs you to Enterprise.
• Included: all Claude models, Claude Code, Cowork, projects with a 200k context window, central billing, SSO, JIT provisioning, role-based permissions, spend controls, enterprise search and workplace connectors.
• Usage: increased limits over individual plans, with premium seats offering significantly more usage than standard seats. The pricing page also states there is no default model training on Team content.
• Watch out for: a premium seat costs five times a standard seat. Mixing seat types is the point, so do not buy premium for everyone because two people need it.

Claude Enterprise.

• Price: USD 20 a seat self-serve, plus usage at standard API rates. Sales-assisted Enterprise is custom priced and quote-only.
• Billing: a single seat type, priced per user per month and billed annually. Self-serve starts at 20 seats, sales-assisted at 50.
• Included: everything in Team plus spend limits, fine-grained role-based access, SCIM, audit logs, a compliance API, custom data retention, network access controls, IP allowlisting and a HIPAA-ready offering.
• Usage: not included. The Anthropic Help Center states that every token your team uses is billed at standard API rates on top of the seat cost.
• Watch out for: this is the only plan where a busy week produces a bigger invoice. Twenty seats at USD 20 is a USD 400 monthly floor with an open meter above it. Set spend limits on day one.

Claude API pricing per million tokens

Claude API pricing is charged per million tokens, separately for input and output, with no seat fee and no monthly minimum. The rates below come from the Claude platform documentation, read on August 28, 2026. MTok means one million tokens, and one token is roughly four characters of English.

• Claude Haiku 4.5: USD 1 per MTok input, USD 5 per MTok output. Batch: USD 0.50 and USD 2.50.
• Claude Sonnet 5: USD 2 per MTok input, USD 10 per MTok output. Batch: USD 1 and USD 5.
• Claude Sonnet 4.6 and Sonnet 4.5: USD 3 per MTok input, USD 15 per MTok output. Batch: USD 1.50 and USD 7.50.
• Claude Opus 5, 4.8, 4.7, 4.6 and 4.5: USD 5 per MTok input, USD 25 per MTok output. Batch: USD 2.50 and USD 12.50.
• Claude Fable 5: USD 10 per MTok input, USD 50 per MTok output. Batch: USD 5 and USD 25.
• Claude Mythos 5: USD 10 per MTok input, USD 50 per MTok output, at limited availability.
• Retired but still billed on partner clouds: Claude Opus 4.1 at USD 15 and USD 75, Claude Haiku 3.5 at USD 0.80 and USD 4.

Two details in that list are easy to miss and both save money. Claude Sonnet pricing of USD 2 and USD 10 launched as introductory pricing due to end on August 31, 2026, and the platform documentation now states it is the standard price and the increase to USD 3 and USD 15 will not happen. Claude Opus pricing is flat at USD 5 and USD 25 across every version from 4.5 to 5, so moving to a newer Opus costs nothing per token. Which model to run is a separate decision from what it costs, and the model-selection lesson in the campus course covers it properly.

Claude API discounts and cost modifiers

Five published modifiers sit on top of the base Claude model pricing, and two are large enough to change which plan you should buy. Most competing pricing roundups mention only the batch discount.

• Prompt caching: a 5-minute cache write costs 1.25x base input, a 1-hour write costs 2x, and a cache read costs 0.1x. Caching pays for itself after one read on the 5-minute duration and two reads on the 1-hour one.
• Batch API: 50% off both input and output tokens for asynchronous work. It stacks with prompt caching, so a cached, batched workload runs far under the headline rate.
• Long context: Claude 4.6 and later include the full 1M token context window at standard pricing. A 900k-token request bills at the same per-token rate as a 9k-token request, so there is no long-context surcharge.
• Tokenizer change: Claude 4.7 and later use a newer tokenizer producing approximately 30% more tokens for the same text, while Sonnet 4.6 and earlier use the previous one. The same document costs more on a newer model at an unchanged rate.
• Data residency: for Claude 4.6 and later, pinning inference to the US applies a 1.1x multiplier on input, output, cache writes and cache reads. Global routing is the default and is charged at standard pricing.

Server-side tools bill separately again. Web search costs USD 10 per 1,000 searches plus the tokens the results consume, web fetch adds nothing beyond tokens, and code execution is free alongside web search or web fetch, otherwise giving each organization 1,550 free hours a month before USD 0.05 per hour per container. Fast mode, in research preview for Claude Opus 5 and Opus 4.8, prices those models at USD 10 and USD 50 per million tokens and cannot be combined with the Batch API.

Claude Code and Claude Cowork pricing

Claude Code and Claude Cowork pricing is simply the plan price: both are included from Pro upward and neither is on Free. Pro at USD 20 a month, both Max tiers, both Team seat types and the Enterprise seat fee all carry them, which makes Pro the cheapest legitimate route into Claude Code rather than the API.

What a plan price does not tell you is what a running agent consumes against a five hour session window and a weekly cap, which behaves nothing like a person typing into a chat box. That is a separate question with a separate answer, and it belongs in a guide on Claude Code economics rather than in a plan comparison. For the setup side, start with the Claude Code guide in the related list below.

Subscription versus API: one month of arithmetic

Claude subscription pricing is flat and API pricing is metered, so the crossover is easier to see with real numbers than with a rule of thumb. Here is one month of a realistic single-person workload priced both ways at the rates checked on August 28, 2026: one operator, 25 working days, ten conversations a day, so 250 conversations, each accumulating about 40,000 input tokens across its turns and producing about 4,000 output tokens.

• Input volume: 250 conversations x 40,000 tokens = 10,000,000 input tokens.
• Output volume: 250 conversations x 4,000 tokens = 1,000,000 output tokens.
• API with Claude Sonnet 5: 10,000,000 x USD 2 per MTok = USD 20.00, plus 1,000,000 x USD 10 per MTok = USD 10.00. Total USD 30.00.
• API with Claude Opus 5: 10,000,000 x USD 5 per MTok = USD 50.00, plus 1,000,000 x USD 25 per MTok = USD 25.00. Total USD 75.00.
• API with Sonnet 5 and prompt caching, 75% of input served as cache reads at 0.1x: 7,500,000 x USD 0.20 = USD 1.50, plus 2,500,000 x USD 2 = USD 5.00, plus USD 10.00 output. Total USD 16.50.
• Claude Pro subscription: USD 20.00 flat, or about USD 16.70 a month on the USD 200 annual plan.

Read the crossover off those six lines. At Sonnet 5 rates, USD 20 of API spend buys 10,000,000 input tokens or 2,000,000 output tokens, so roughly ten conversations a day lands within a few dollars of the Pro price either way. Below that volume the API is cheaper and you should not subscribe. Above it, or on Opus 5 where the same month costs USD 75, the subscription wins clearly, because a flat fee absorbs a bad week that a meter would not. The exception is well-cached programmatic usage: at USD 16.50 for identical tokens, caching drags the API back under the subscription and keeps it there as volume grows.

Which Claude plan should you pick

Three factors decide this: volume shape, whether a program or a person generates the requests, and whether anyone needs to administer the account. Everything else is noise. Four concrete situations, one recommendation each.

• A few sessions a week for writing, research and thinking: stay on Free at USD 0. You will notice the usage ceiling before you notice a missing feature, and that is your upgrade signal.
• One founder using Claude daily who wants Claude Code and Cowork: buy Pro at USD 20 a month, or USD 200 for the year. It is the lowest rung that unlocks those tools, and annual saves about USD 40.
• Hitting the five hour session limit most working days on Pro: move to Max 5x at USD 100 a month, not Max 20x at USD 200. The tiers ship identical features, so the extra USD 1,200 a year buys unproven capacity.
• Two to five people needing shared billing, SSO and admin controls: Team at USD 20 per member a month annually, with USD 100 premium seats only for the one or two who genuinely exhaust a standard seat.
• Building a product on Claude rather than chatting with it: skip subscriptions for the API, where Haiku 4.5 is USD 1 and USD 5 and Sonnet 5 is USD 2 and USD 10, then turn on prompt caching and the Batch API before optimizing anything else.

Two negatives deserve stating as plainly as the recommendations. Do not buy Max 20x at USD 200 a month before spending a full month on Max 5x at USD 100 and hitting its weekly limit repeatedly, because the tiers differ only in capacity. And do not pick Enterprise self-serve on the strength of its USD 20 seat price: usage bills at API rates on top, so 20 seats is a USD 400 monthly floor with an open meter above it, and it earns its place only when you actually need SCIM, audit logs and IP allowlisting.

Re-check every figure here before you buy, because Anthropic changes them regularly. If you want the reasoning behind the token math rather than the totals, the free courses and glossary at Agentic School cover context windows, prompt caching and cost estimation end to end.

Frequently asked questions

Keep learning