Enterprise AI Economics

Claude Code Pricing in 2026: Pro, Max and Ultra Limits, API Cost and Team Seats

Claude Code has no price of its own. You pay for it through a Claude subscription (Pro, Max or Ultra), a Team or Enterprise seat, or API tokens. This guide sets out each route as Anthropic published it on 14 November 2026: plan prices, the five-hour and weekly limits, token prices, usage credits, what 10, 50 and 200 developers pay in a year, and when a local model is the cheaper engine.

Claude Code pricing has no separate line: the tool is included in Pro at $20 a month ($17 on the prepaid year), Max at $100 and Ultra at $200, and in Team seats from $20 a month, with usage capped by five-hour or weekly limits (Anthropic, 14 November 2026). With an API key you pay per token, from $2 per million input tokens on Sonnet 5.5.

Anthropic sells Claude Code three ways. A personal subscription bundles it with the Claude apps and caps it with usage windows. A Team or Enterprise seat adds central billing and admin controls. A Claude Console account, or a cloud provider such as Amazon Bedrock, bills every token with no cap except the one you set. Which route is cheapest depends almost entirely on how many hours a day a developer keeps an agent running.

The prices, limits and quotations in this guide were read on 14 November 2026 from Anthropic’s pricing page, the Claude Help Center, the Claude Code documentation and the Claude Platform price list. Company-wide seat economics, including SCIM, audit logs and the annual commitment, are covered in our separate breakdown of Anthropic’s enterprise seat model; this page stays with the coding tool.

Claude Code pricing: every way to pay

Route (US list, verified 14 November 2026)PriceWho it is forClaude Code usageAdmin features
Free$0Trying Claude chatNot includedNone
Pro$20 a month, or $17 a month as $200 billed up front for a yearOne developer, occasional sessions1x allowance on a five-hour and a weekly limit; Fable only through usage creditsNone; training on your content is opt-out
Max$100 a month, monthly billing onlyAll-day individual use5x Pro on both limits; Fable up to half the weekly limitNone; $100 a month of API credits
Ultra$200 a month, monthly billing onlyThe longest-running tasks10x Pro each week; no five-hour limit for verified accountsNone; $200 a month of API credits
Team, Standard seat$20 per seat a month billed annually; $25 billed monthlyTeams of 2 to 1501.25x Pro’s per-session allowance, plus a weekly limitCentral billing, SSO, admin controls, no training by default
Team, Premium seat$100 billed annually; $125 billed monthlyHeavier users on the same team6.25x Pro’s per-session allowanceAs Standard; seat types can be mixed
EnterpriseUS$20 per seat a month, billed annually, 20-seat minimumLarger organisationsUsage-based plan: no per-seat limit, every token billed at API rates; seat-based Enterprise plans have included limitsSCIM, audit logs, Compliance API, spend limits per user and organisation
Claude Console API keyNo subscriptionAutomation, CI, teams that prefer metered billingUnlimited within your rate limits; billed per tokenWorkspace spend limits, per-user analytics
Amazon Bedrock, Google Cloud, Microsoft FoundryBilled by the cloud providerOrganisations buying through a cloud accountPer token on the cloud invoiceThe cloud’s own budget controls

Anthropic’s prices exclude tax. Max and Ultra cannot be bought annually, and Pro is the only personal plan with a prepaid discount. The Enterprise row is included for completeness, because on the usage-based Enterprise plan the pricing page lists, Claude Code usage is charged the same way as an API key.

Is Claude Code free?

No. The installer is free to download, but Claude Code will not answer a prompt until it has a paying account behind it. Anthropic’s plan comparison puts Claude Code in the Pro column and above, and the authentication page in the Claude Code documentation lists the accounts it accepts: a Pro, Max or Ultra subscription, a Team or Enterprise seat, a Claude Console login, or credentials for a cloud provider. The Free plan is not on that list.

Two details catch people out:

  • An API key overrides your subscription. If ANTHROPIC_API_KEY is set in your shell and you approve it, Claude Code bills that key per token, even though you are paying for Pro or Max. Run unset ANTHROPIC_API_KEY, then check /status.
  • The monthly API credits on Max, Ultra and Team do not pay for interactive Claude Code. They cover the API, the Console, Managed Agents and the Agent SDK. They also cover headless claude -p runs, but only when those runs use an API key from the linked Console organisation.

Claude Code limits: the five-hour window and the weekly cap

Subscriptions meter Claude Code with usage windows, not with a token counter you can read off a price list.

  • Five-hour session limit. Applies on Pro, Max and Team seats. When it is reached, Claude Code shows the reset time and waits. Anthropic is rolling out a small wrap-up allowance that lets the current step finish; it counts against the weekly limit.
  • Weekly limit. Applies on every subscription, Ultra included. The Team plan article says it resets at a fixed time assigned to the account, visible in Settings, Usage.
  • A separate weekly limit for Fable models on Max and Ultra.
  • One allowance for everything. The limits are shared between the Claude apps and Claude Code, so a morning of long chats in the browser shortens the afternoon in the terminal.

Anthropic publishes the limits as multiples of Pro and gives no token or message figures. Its pricing page says how fast you reach a limit depends on the model, the length of the conversation and the size of the files involved. Model choice matters here as much as on the API: the default model on every plan is Opus 5.5, which the documentation describes as costing more than Sonnet, and a model-specific message such as “You’ve hit your Opus limit” can appear before the general limit does. Switching to another model family with /model keeps the session going in that case.

Inside Claude Code, /usage shows the plan’s usage bars, a breakdown of what consumed the allowance (skills, subagents, plugins, MCP servers) and suggested changes. The documentation is frank about why long sessions burn through limits: the whole conversation is re-sent with every request, a break longer than the prompt-cache lifetime forces the context to be reprocessed, and agent teams use roughly seven times the tokens of a standard session when teammates run in plan mode.

Claude Code Pro vs Max vs Ultra

What changes (Anthropic pricing page, 14 November 2026)ProMaxUltra
Monthly price$20, or $17 on the prepaid year$100$200
Five-hour limit1x5xNone for verified accounts
Weekly limit1x5x10x
Fable modelsPaid through usage creditsIncluded, up to 50% of the weekly limitIncluded, up to 50% of the weekly limit
Opus, Sonnet, HaikuYesYesYes
Monthly API creditsNone$100$200
Priority access at busy timesNoYesYes
Usage credits when a limit is reachedYesYesYes

The step from Pro to Max is a fivefold price for a fivefold allowance, plus Fable and API credits. The step from Max to Ultra doubles the price, doubles the weekly allowance and removes the five-hour window, which is the limit that interrupts long autonomous runs. A developer who rarely hits a limit on Pro gains nothing from Max. One who hits the five-hour limit daily usually finds Max cheaper than topping up with usage credits at API rates.

All three are personal plans. Model training is opt-out, there is no single sign-on and no central billing, so they are a poor fit for company code even when the arithmetic looks attractive. What each plan does with prompts and files is set out in our guide to Claude and confidential data.

Claude Code API cost: what a token costs

With a Console API key, usage credits or an Enterprise seat, Claude Code is charged at Claude Platform list prices.

Model (Claude Platform list, 14 November 2026)Input, per million tokensOutput, per million tokensCache read, per million tokens
Claude Fable 5.2$8$40$0.20
Claude Opus 5.5, the Claude Code default$4$20$0.20
Claude Sonnet 5.5$2$10$0.10
Claude Haiku 5.5, prompts up to 100,000 tokens$0.10$0.50$0.01
Claude Haiku 5.5, prompts over 100,000 tokens$0.50$2.50$0.05

Three multipliers sit on top. US-only inference costs 1.1 times the list price. Extended thinking is billed as output tokens and cannot be switched off on the 5.5 and Fable models. And the tokenizer used from Claude 4.7 onward produces roughly 30% more tokens for the same text than the previous one, by Anthropic’s own estimate.

Anthropic’s cost documentation gives the only vendor-published planning figure: across enterprise deployments, Claude Code averages about $13 per developer per active day and $150 to $250 per developer per month, with 90% of users below $30 per active day. It recommends a pilot group to set your own baseline, and for API organisations it suggests rate limits of 200,000 to 300,000 tokens per minute per user for teams of up to five, falling to 10,000 to 15,000 for teams above 500.

Usage credits and monthly API credits are different things

  • Usage credits let a Pro, Max, Ultra or Team member keep working after a limit. They are prepaid and billed at standard API rates, with optional auto-reload and a monthly spend cap you set. Credits bought on or after 7 December 2026 expire 12 months after purchase.
  • Monthly API credits are a plan benefit: $100 on Max, $200 on Ultra, and $20 per Standard or $100 per Premium Team seat, pooled and capped at $500 a month. They expire each billing cycle and cannot be spent in interactive Claude Code or as usage credits.

Claude Code cost for 10, 50 and 200 developers

The table multiplies list prices by 12 months. The usage rows use Anthropic’s published average of $150 to $250 per developer per month, which is the vendor’s figure for enterprise deployments and no substitute for your own pilot data.

Route (US list, 14 November 2026)Per developer per year10 developers50 developers200 developers
Team, Standard seats, billed annually12 × $20 = $240$2,400$12,000Not offered: Team stops at 150 seats
Team, Standard seats, billed monthly12 × $25 = $300$3,000$15,000Not offered
Team, Premium seats, billed annually12 × $100 = $1,200$12,000$60,000Not offered
Team, four Standard seats to one Premium, annualMixed8 × $240 + 2 × $1,200 = $4,32040 × $240 + 10 × $1,200 = $21,600Not offered
Enterprise seats, before usage$240$4,800, because 20 seats is the minimum$12,000$48,000
Enterprise usage at $150 to $250 a month$1,800 to $3,000$18,000 to $30,000$90,000 to $150,000$360,000 to $600,000
Enterprise, seats plus usage$2,040 to $3,240$22,800 to $34,800$102,000 to $162,000$408,000 to $648,000
Console API keys, no seats, same averages$1,800 to $3,000$18,000 to $30,000$90,000 to $150,000$360,000 to $600,000

How to read it:

  • Team is the cheap route while it lasts. A Standard seat costs less in a year than Anthropic’s average metered usage costs in two months, because the limits cap what a member can consume.
  • Past 150 people, every route is metered. At 200 developers the only choices are Enterprise seats with usage on top, API keys, or a cloud provider.
  • Seat mix is the main lever on Team. Moving one developer in five to Premium nearly doubles the bill for 50 people, from $12,000 to $21,600.

The cost lines a seat price hides

  • Usage credits. Once switched on, every hour past a limit is billed at API rates. Set the organisation and per-member caps before the first developer hits a wall.
  • Premium model multipliers. Fable 5.2 costs twice what Opus 5.5 does per token and four times Sonnet 5.5. Anthropic’s documentation traces unexpectedly high API spend to two habits: long sessions that were never cleared, and Opus left as the default for routine edits.
  • Seats that sit idle. A subscription seat costs the same whether its owner opens Claude Code daily or once a month, and Enterprise seats are billed for the full year.
  • API fallback. A stray ANTHROPIC_API_KEY, a CI job or a headless script bills per token outside the plan.
  • Agent teams, subagents and scheduled tasks. Each one sends its own requests, and scheduled tasks keep firing while a session is idle.

The same exercise for other tools is in our guides to what Cursor charges, OpenAI’s Codex plans and GitHub Copilot’s credit-based billing.

When your own hardware is cheaper than a Claude Code plan, and when it is not

Claude Code can be pointed at another endpoint with ANTHROPIC_BASE_URL, and local servers now speak the Anthropic message format, so a model on your own hardware can drive the same agent loop with no token bill from Anthropic. Our set-up guide for Claude Code with Ollama, LM Studio and llama.cpp has the exact variables. Anthropic’s gateway documentation says plainly that routing Claude Code to non-Claude models is not supported, so this is a do-it-yourself arrangement.

A local model tends to be cheaper when:

  • You already hit limits every day. Heavy users are the ones paying for Premium seats, Ultra or usage credits. Moving routine work (tests, refactors, boilerplate) to a local model and keeping Claude for the hard problems cuts the metered part.
  • The hardware exists. A workstation or shared GPU server that is already paid for makes the marginal cost of a request electricity.
  • Many developers share one server. Utilisation is what lowers the cost per request on owned hardware.
  • The code may not leave the building. Then the comparison is no longer about price.

It tends not to be cheaper when:

  • Use is light. A Standard Team seat at $240 a year is hard to beat with any hardware purchase.
  • You would buy GPUs for one or two people. Idle hardware costs the same as busy hardware.
  • The work needs frontier-model quality. Open-weight models are not Claude. If a weaker model needs three attempts where Opus needs one, the saving goes into developer hours.
  • Nobody has time to run it. Model updates, context settings and tool-calling quirks are now your job.

We do not quote hardware prices here because they move monthly and depend on what you already run. For the method, use our on-premises cost and TCO guide and the explanation of flat versus per-token pricing. For the model itself, see the comparison of open-weight coding models, check memory needs in the VRAM calculator, and compare other harnesses in our roundup of on-prem coding assistants.

How VDF AI fits

VDF AI does not run, host, resell or integrate with Claude Code, and nothing on this page is a VDF price. The honest distinction is between one developer tuning a personal plan and an organisation that has to decide who hosts the model, which models are approved, who can use them on which repositories, and what record exists afterwards.

VDF Code is VDF AI’s governed alternative for that second case: an on-premises coding assistant with inline completions, chat, edit-mode refactors and pull request review, running in your VPC, your data centre or fully air-gapped, on models you approve. Its policy engine can allow or block models and restrict tools by repository, and prompts, retrievals and completions are logged for review. On-premises it is licensed per seat and the licence does not meter model tokens; the pricing page explains the structure.

Sources

Every page below was opened on 14 November 2026; plan names, prices and limits are quoted as they stood that day.


Working out what a coding assistant will cost your team next year? See how VDF Code runs on infrastructure you control, or talk to us about a governed set-up for your developers.

Frequently asked questions

How much does Claude Code cost?

Claude Code is not sold separately. It is included in Claude Pro at $20 a month, or $17 a month when $200 is paid up front for a year, in Max at $100 a month and in Ultra at $200 a month (Anthropic's pricing page, 14 November 2026). Team seats cost $20 or $100 per month on annual billing and also include it. The alternative is an API key from the Claude Console, where each token is billed at list price with no subscription.

Is Claude Code free?

No. Installing the command-line tool costs nothing, but it needs an account that pays for the model. Anthropic's pricing page lists Claude Code from the Pro plan upward and not on the Free plan, and its authentication page accepts Pro, Max, Ultra, Team, Enterprise, Claude Console and cloud-provider accounts. The only way to run it without paying Anthropic for tokens is to point it at a model you host yourself, which Anthropic does not support.

What are the Claude Code limits on Pro, Max and Ultra?

Pro and Max have two limits that apply together: a five-hour session limit and a weekly limit. Ultra has a weekly limit only, and verified Ultra accounts have no five-hour limit. Max gives five times Pro's allowance on both windows; Ultra gives ten times Pro's weekly allowance. The limits are shared with the Claude apps, and Anthropic publishes them as multiples, not as token or message counts (checked 14 November 2026).

Claude Code Pro vs Max: which plan do I need?

Pro at $17 to $20 a month suits occasional sessions: one allowance on a five-hour and a weekly window, with Fable models available only through paid usage credits. Max at $100 a month multiplies both windows by five, includes Fable up to half the weekly limit, adds priority access at busy times and brings $100 a month of API credits. If you reach the five-hour limit most working days on Pro, Max is the plan Anthropic aims at all-day use.

How much does Claude Code cost through the API?

With a Console API key, Claude Code is billed per token at list prices: Opus 5.5 at $4 per million input tokens and $20 per million output tokens, Sonnet 5.5 at $2 and $10, Haiku 5.5 at $0.10 and $0.50 for prompts up to 100,000 tokens, and Fable 5.2 at $8 and $40 (14 November 2026). Anthropic's own documentation reports an average of about $13 per developer per active day and $150 to $250 per developer per month across enterprise deployments.

Can self-hosting a model lower what Claude Code costs?

Technically yes. Claude Code reads the ANTHROPIC_BASE_URL variable, and servers such as Ollama, LM Studio and llama.cpp expose an Anthropic-style endpoint it can talk to, so no tokens are billed by Anthropic. Anthropic states that it does not support routing Claude Code to non-Claude models, so you carry the support burden yourself. Savings depend on hardware you already own and on whether an open-weight model is good enough for the work.

Filed under
AI coding assistantsAI platform pricingAI platform TCOAI procurementlocal LLMon-premises AI
AI Cost & Energy

Calculate your AI infrastructure savings

Model the cost and energy impact of running AI on-prem versus cloud-only — then see the benchmark data behind the numbers.

Or start free — no credit card →

Keep reading