What Is Claude AI? The Models, Plans and Surfaces Anthropic Actually Publishes

Anthropic answers the question on its own site, in the FAQ under the pricing table: Claude is an artificial intelligence, trained by Anthropic using Constitutional AI to be safe, accurate and secure, and in the company's words the trusted assistant for you to do your best work. That sentence settles what Claude is. It does not settle which Claude somebody is looking at, because the same site lists seven products, nine capabilities and four model families, and the names overlap in ways the products do not. This page is the map: what Anthropic publishes, what each individual plan includes, and the one figure Anthropic does not publish. It returns to Claude Opus 5.5 whenever a concrete price is needed, because that is the model whose list price the pricing page states plainly.

Part of the confusion is that two questions share the same three words. One person wants to know where to type; another wants to know what to call in code. Anthropic sells both, on different pages and in different units — a subscription on one side, tokens on the other — and neither page explains the other.

The definition, and the five plans behind it

The rest of that FAQ is as useful as the first sentence. It says you can use Claude for personal work or create a Team account to collaborate with teammates, and it counts the plans plainly: Claude has five pricing plans available, named Free, Pro, Max, Team and Enterprise. That is the count Anthropic gives a newcomer, and it is accurate. It is just not the way the pricing page lays them out.

On claude.com/pricing the five arrive in two groups. Free, Pro and Max sit side by side as individual plans, each with a one-line description of who it is for: Free for everyone, Pro for everyday work, Max for people who work with Claude all day. Team and Enterprise sit on their own track, priced per seat and aimed at groups rather than at a person deciding what to try. Read the two pages together and the arithmetic reconciles: three individual plans plus two organisational ones is five.

Everything else on the site hangs off those two axes. A product is a place you go. A plan is what you are allowed to do when you get there. A model is what answers. Most misunderstandings about Claude come from sliding one of those onto another.

Seven products and nine capabilities

Anthropic's product navigation lists five products — Claude, Claude Code, Claude Cowork, Claude Duet and Claude Tag — and, under a heading called Specialized, two more: Claude Security and Claude Science. That is seven destinations. Three of the five main ones are hard to place from the name alone, which is why the names are worth spelling out rather than paraphrasing: Cowork, Duet and Tag are products, not features, and each has its own entry in the menu.

The capabilities list is longer and easier to mistake for a product list. It contains Artifacts, Design, Connectors, Plugins, Skills, Extensions, Claude in Chrome, Claude for Google Workspace and Claude for Microsoft 365. Nine items, none of them something you go to on its own. They are things Claude does, or places Claude appears inside software somebody else made. The distinction is Anthropic's, not ours, and it is not a ranking: Design appears under Capabilities in the menu while Claude Design is included on the Free plan, because a capability can be sold inside a product without being one.

That menu is also visibly in motion. The pricing page carries a banner reading "Claude Cowork is now just Claude. Rolling out to Pro and Max, with more plans to follow," while the navigation still lists both names. A banner describing a rollout is not yet a finished product list, and this page treats it that way: the two names coexist on the site as of 23 October 2026, and the direction of travel is a merge.

What the front door sells is continuity: one sign-in, and the conversation is on the web, in the desktop app and on the phone. That is the offer the Free plan lists first, and it is why seven products can share one menu.

Four model families, and what a family name buys you

The models column of the menu has four entries: Mythos, which includes Fable; Opus; Sonnet; and Haiku. Those are family names, and a family name is a position in a line-up rather than a promise about capability. Each family carries a version, and the version is what you actually select. The pricing page's model table, read on 23 October 2026, lists Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 5.5 — the same four families, each with a number attached.

The plan grid is where the families stop being abstract, because it shows which ones a plan includes. Sonnet and Haiku are marked included on every plan including Free. Opus is marked No on Free and Yes from Pro upward. Fable gets a row of its own, marked No on Free, with a note about usage credits and the footnote that usage limits apply. If you are on Free and a page asks you to switch models, that is what you are choosing between: two of the four families, not all of them.

The platform table, a little further down the same page, is where Anthropic states prices without any relative language at all. Fable 5.1 is listed at $10 input and $50 output per million tokens. Opus 5.5 is $4 and $20. Sonnet 5.5 is $2 and $10. Haiku 5.5 is priced in two tiers: $0.10 input and $0.50 output per million tokens for prompts up to 100,000 tokens, and $0.50 and $2.50 above that. Cache reads carry their own rates, from $0.20 per million tokens on Opus 5.5 down to a cent on the small Haiku tier. The context window is described the same way in the plan grid as in the platform table: up to 1M, varying by model.

That contrast matters more than any number in it. The same page will tell you a million tokens of Opus 5.5 costs exactly four dollars in, and will not tell you how many tokens twenty dollars buys on Pro. The published half is precise. The unpublished half is the subscription.

What Free, Pro and Max each include

Free is $0 and is described as free for everyone. It covers chat on web, desktop and mobile; searching the web, creating files and running code; memory across conversations; connecting your apps and tools; and Artifacts, Claude Design, Claude Slides and Claude Docs. It also caps Projects at five, which is the one place on the individual grid where a plan difference is expressed as a count rather than a comparison.

Pro is $17 a month with annual billing, billed as $200 up front, or $20 if billed monthly. It carries everything in Free and adds more usage, hand-off and scheduling of tasks, Claude Dashboards in beta, Claude Code, Claude Science, Projects, access to more Claude models, Claude in Chrome and Claude for Microsoft 365. For most people the meaningful additions are the two that change what Claude can reach: Claude Code, and the Chrome and Microsoft 365 integrations, which put it inside software that is not Anthropic's.

Max starts at $100 a month and is the plan for people who work with Claude all day. It includes everything in Pro and adds a choice of 5x or 20x more usage than Pro, higher output limits on all tasks, early access to advanced features, priority access at high traffic times, and between $100 and $200 a month in API credits to build on the Claude Platform. Two of those are worth separating from the rest. Priority access is about queueing, not capacity. The API credits are a different currency from the plan itself, and the next section is about why that distinction is the honest one.

The figure Anthropic does not publish

No page on claude.com states a token allowance for any individual plan. The pricing page quantifies usage only in relation to another plan. Pro's addition over Free is described as more usage. Max's addition over Pro is described as 5x or 20x more usage than Pro, depending on which tier you buy. Free is not quantified at all. There is no message allowance, no monthly token pool and no credit balance published for the subscription tiers.

Anthropic's own footnote is explicit about the terms and silent about the quantity: "Usage limits apply. Prices shown don't include applicable tax. Price and plans are subject to change at Anthropic's discretion." That sentence appears twice on the page. It is the closest the pricing page comes to describing what a limit even is.

What this means practically is that "how much Claude do I get for $20" has no answer on Anthropic's site, and any third-party figure offered as an answer is a measurement somebody took, not a specification Anthropic published. The multipliers are real information — Max at 20x does deliver twenty times the usage of Pro by Anthropic's own account — but a multiplier of an unpublished base is not a quantity. Treat "more usage" as a direction, not a budget.

There is exactly one bridge between the two currencies, and it is on Max. The $100 to $200 a month in Claude Platform credits is spent at the published per-token rates, so those credits convert to tokens at prices anybody can read. Nothing else in the individual plans does. If you need a number before you commit, the platform table gives you one and the plan grid does not, and that asymmetry is a fact about what Anthropic chose to publish rather than a gap in this page's research.

Team and Enterprise are a different purchase

Anthropic keeps the organisational plans on a separate track from the three individual ones, priced per seat, with Enterprise carrying the option of a conversation with a buying specialist. The page frames Team for teams of 2 to 150 and Enterprise for large businesses operating at scale, and the feature comparison for those plans runs to single sign-on, audit logs, role-based access, custom data retention and a HIPAA-ready offering — none of which appears in the individual grid at all.

The practical point for a newcomer is narrower: the individual plans are the ones you can compare on price today, and Team and Enterprise are bought for a group, by someone with a budget and an IT department. They are not a fourth step after Max. If the question is what Claude costs a person, three plans are the whole answer.

The subscription and the API are different products

The second commonest confusion, after which Claude is, is why a paid plan still asks for an API key. It does because a plan buys a seat in Anthropic's surfaces and an API key buys calls to a model from software Anthropic did not write. The two are billed differently, limited differently and consumed by different people. A Pro subscription does not arrive with a token balance, and a platform key does not unlock Claude Code. The only place they touch is the API credit bundled into Max, which is a platform allowance attached to a subscription rather than the subscription itself.

If your work is the second kind — software calling a model — the plan is not the thing to compare. OrcaRouter hosts models, not Anthropic's apps or anyone else's assistant, so a Claude subscription is not something our endpoint replaces. What the endpoint does is remove the plumbing: one OpenAI-compatible interface for 200-plus models behind one key, at provider list price with no markup, with automatic failover when a provider has a bad minute. Claude Opus 5.5 runs there at $4.00 input and $20.00 output per million tokens, with a 1M-token context and cache reads at $0.20 per million.

The model table is a price list for the platform, not for the subscription. The same four family names appear on both, which is what makes the two look interchangeable when they are not.

If you are new, three questions in order

First, which surface. If you want to type at an assistant that remembers your conversations, that is Claude, and Free covers it. If you want the product Anthropic files under Developers, that is Claude Code, and it begins at Pro. If your work is security or scientific research, the Specialized entries are the ones to read before the general ones.

Second, which plan. Free is a genuine free tier rather than a trial, and its limits are the five-project cap and the two model families. Pro is the point at which the surfaces widen: Claude Code, Projects, the browser and Microsoft 365 integrations, and Opus rather than only Sonnet and Haiku. Max is a capacity purchase, and the only reason to make it is that you already know you will use more than Pro allows.

Third, which model, and only when it matters. Pick the family first — Opus for the top of the line, Sonnet for the middle, Haiku for volume, Fable on its own footing — and then the version, because a family name without a version is not a model you can select. On Free the grid has already chosen; everywhere else the version is the thing to check before you rely on it.

The menu will keep moving; the banner about Cowork is this month's proof. What is unlikely to change quickly is the shape underneath it: a short product list, a longer capability list, four model families, three individual plans, and a usage figure that stays relative. Learn the shape once and the next rename costs you nothing.