Claude Code Pricing in 2026: What Each Tier Actually Costs

What Claude Code pricing actually covers

There is no standalone Claude Code price. Claude Code is packaged with a Claude subscription, and the number on your card is the subscription number. On Anthropic's pricing page (checked 18 September 2026), Claude Code first appears at the Pro tier. The Free plan does not include it.

That single fact causes most of the confusion around this term. People search for the price of a developer tool and find the price of a chat subscription, because they are the same thing.

One distinction to get out of the way early, because it decides which page you actually want. This post is about the subscription tiers. If you are building on the Anthropic API and want the per-token rate card for input, output, and cached reads, that is a different billing system with different numbers, and it is covered in Claude API pricing. The two only intersect at one point, covered further down: when you exhaust a subscription limit, overflow is billed at API rates.

Claude Code pricing at a glance

Sources: claude.com/pricing, What is the Pro plan?, What is the Max plan?, What is the Team plan?, and What is the Enterprise plan?, all checked 18 September 2026.

Four details in that table are easy to miss.

Max bills monthly only, so the annual discount that saves roughly 15% on Pro is not available on the tier most heavy users end up on. Anthropic's pricing FAQ is explicit that annual subscriptions exist for Pro and Team, and that Enterprise is annual only.

Team seats come in two very different sizes. A Standard seat is barely above Pro in capacity, while a Premium seat sits between Max 5x and Max 20x. Anthropic's Team article puts the multipliers at 1.25x and 6.25x of Pro's per-session allowance.

Enterprise now carries a published seat price. The pricing page lists $20 per seat per month, billed annually, plus usage billed at API rates, and says you can start self-serve or go through sales for a tailored quote. That is a change worth noting if you are working from a quote or an article written earlier in 2026, when Enterprise was listed only as custom pricing.

And the multipliers are per session, not per month. That distinction is doing a lot of work, and it is the subject of the next section and of an active lawsuit covered further down.

How the usage limits actually work

Every paid plan runs two clocks at once. There is a rolling five-hour session window, and on top of that a weekly limit that resets at a fixed time assigned to your account. Anthropic documents both on the Pro plan help page and repeats the structure for Max. Your reset day and time stay the same regardless of when you subscribed, and you get the full weekly allowance each cycle.

The multipliers you see advertised apply to the session window. Max 5x is five times Pro's per-session allowance, Max 20x is twenty times. The weekly ceiling is a separate constraint and does not scale by the same factor, which is the single most common misreading of the pricing page.

There is also a third clock that almost nobody budgets for. Anthropic's usage limit best practices article (checked 18 September 2026) describes the Usage settings page as showing your weekly reset "for Opus only and all other models," so the weekly allowance is really two allowances: one that all models draw from, and a separate one specific to Opus. A week of heavy Opus work can leave you with Sonnet capacity and no Opus capacity, which reads like a bug and is not one. Anthropic does not publish the ratio between the two, and the cost documentation confirms the practical consequence: after a model-specific limit message, switching to another model family keeps you working.

Three more mechanics matter for budgeting.

Claude Code and Claude chat draw from the same allowance. Anthropic's usage and length limits article states that usage across all Claude surfaces, claude.ai, Claude Code, and Claude Desktop, counts toward the same usage limit. The usage credits documentation (published 10 August 2026) says the same about combined usage. A long planning conversation in the morning reduces what is left for the agent in the afternoon. IDE usage counts too: Anthropic's Pro and Max guide for Claude Code confirms your seat covers Claude Code in VS Code, Cursor and other VS Code forks, and JetBrains IDEs, all metered against the same pool.

Hitting the cap does not have to stop work, but the escape hatch is not free. With usage credits enabled, you can keep going past the included limit and the overflow is billed at standard API rates. That is where the subscription and the API rate card meet. Without credits enabled, you wait for the window to reset.

Which limit you hit changes what you can do about it. Anthropic's cost management documentation draws the distinction precisely: a session or weekly limit message means a plan-wide window shared across all models, so switching models with `/model` will not restore access, while a model-specific message such as "You've hit your Opus limit" does leave other model families available. Knowing which message you are looking at saves a support ticket.

Anthropic does not publish a token quota per tier. Several third-party posts do, quoting figures like a fixed token count per five-hour window. Treat those as estimates from individual measurement rather than vendor-published numbers, because the vendor's own pages describe capacity only as a multiple of Pro. Where a third-party figure conflicts with the live pricing page, the pricing page wins.

The closest thing to an official hours figure came from the launch of weekly limits. When Anthropic announced them, it told TechCrunch (28 July 2025) that most Pro users could expect 40 to 80 hours of Sonnet 4 per week, Max 5x subscribers 140 to 280 hours of Sonnet 4 plus 15 to 35 hours of Opus 4, and Max 20x subscribers 240 to 480 hours of Sonnet 4 plus 24 to 40 hours of Opus 4. Those numbers are from 2025 and name models that have since been replaced, so do not plan against them. They are useful for one thing only: showing that the ratio between tiers in practice was never a clean 5x and 20x, because the Opus allowance scaled far more slowly than the Sonnet one.

What a month actually costs: four worked scenarios

The figures below are arithmetic, not survey data. Every assumption is stated so you can substitute your own and rerun it. The per-token benchmark comes from Anthropic's own cost documentation, which reports an average of around $13 per developer per active day and $150 to $250 per developer per month across enterprise deployments, with 90% of users below $30 per active day. The plan prices come from claude.com/pricing.

Scenario one: the occasional user. Assumption: Claude Code a few times a week, short sessions, mostly Sonnet, plus normal chat use. Say six active days a month at the low end of Anthropic's range, around $5 per active day, so roughly $30 of equivalent token consumption. Pro at $20 per month, or $17 on annual billing, covers this comfortably and leaves chat capacity. There is no case for Max here, and the Free plan is not an option because it does not include Claude Code at all.

Scenario two: the average working developer. Assumption: Claude Code on most working days, roughly 18 active days a month, at Anthropic's stated $13 average. That is about $234 of equivalent consumption, which sits inside Anthropic's own $150 to $250 per developer per month band. Against subscription prices, $234 is more than Max 5x at $100 and slightly more than Max 20x at $200. The practical read is that this developer is a Max 5x buyer who should watch for weekly limit messages, and upgrades to Max 20x only if those messages arrive more than once or twice a month.

Scenario three: the heavy user. Assumption: 20 active days, longer sessions, Opus for planning, several parallel agents. Anthropic's own documentation says 90% of users stay under $30 per active day, so take $30 as the heavy-user ceiling: 20 times $30 is $600 of equivalent consumption against a $200 subscription. Max 20x is the obvious plan, and the interesting question becomes what happens at the ceiling rather than what the plan costs. If this developer runs out of weekly capacity on a Thursday twice a month, the honest options are usage credits at API rates, usage bundles at a discount, or moving that workload to an API key.

Scenario four: a ten-person engineering team. Assumption: seven developers use Claude Code moderately and three run it hard, billed annually. Seven Standard seats at $20 plus three Premium seats at $100 is $140 plus $300, or $440 per month, against $550 per month at monthly rates. Now compare that with the same ten people on usage-based Enterprise: $20 per seat per month is $200 in seat fees, plus every token at API rates. Using Anthropic's own $150 to $250 per developer per month figure, that same team consumes $1,500 to $2,500 of usage on top, so $1,700 to $2,700 a month. The Team plan is dramatically cheaper for a team of this size and shape, and the reason is simple: Team seats include usage, Enterprise seats do not. Enterprise buys governance, not capacity.

One caveat that applies to all four. Anthropic's cost documentation notes that per-developer costs vary widely with model selection, codebase size, and usage patterns such as running multiple instances or automation, and recommends starting with a small pilot group to establish a baseline before wider rollout. That advice is worth more than any table, including this one.

Subscription vs API billing, and where the crossover sits

Anthropic's own cost management documentation gives the most useful public number here. Across enterprise deployments, it reports an average of roughly $13 per developer per active day and $150 to $250 per developer per month, with 90% of users staying below $30 per active day. Those figures are measured at API token rates.

Set that against the flat subscription prices and the crossover becomes simple arithmetic. At $13 per active day:

• About 8 active days of typical use equals the $100 Max 5x price.
• About 15 active days equals the $200 Max 20x price.

So anyone using Claude Code most working days is consuming more at API rates than the top individual subscription costs. Finout's 2026 cost analysis reaches the same range independently, putting typical Claude Code cost at scale at $150 to $250 per developer per month.

The honest counterweight is that API billing has no cap and no weekly quota. A team with spiky, unpredictable usage, or one that needs to run automation without a human session attached, may reasonably prefer per-token billing even when it costs more on average, because it never blocks. A subscription optimises for predictable cost. A key optimises for never waiting.

Claude Code vs Cursor, Windsurf, Copilot, and Codex

Sticker price is the least useful column in this comparison, because every vendor now sells a quota and the quotas are not comparable units. The table below pairs each price with what the quota is actually denominated in, which is the part that decides your bill.

Sources, all checked 18 September 2026: claude.com/pricing, Cursor's pricing page and Cursor's pricing docs, Devin's pricing page (windsurf.com/pricing now returns a 308 redirect there), GitHub's Copilot plans documentation, and OpenAI's Codex pricing page.

Three differences matter more than the prices.

Claude Code is the only one of these that meters time rather than money or credits. Cursor gives you a pool denominated in dollars and drains it at the underlying model's API price. Copilot gives you a credit balance where one credit is one cent, with $10 of base credits on Pro and $100 on Max. Claude gives you a session window and a weekly window, and never tells you the token number behind either. That makes Claude harder to forecast in advance and easier to live with day to day, because you cannot accidentally spend $80 in an afternoon without opting in to usage credits first.

Copilot pools credits across an organisation. Claude Team does not. GitHub's documentation says each license contributes AI credits to a shared enterprise pool with usage beyond the pool charged at $0.01 per credit. Anthropic's Team article says the opposite in plain terms: usage limits on Team plans are per-member, so one person exhausting their seat does not affect anyone else's. Neither model is better, but they fail differently. Pooling means one heavy user can drain the team; per-member limits mean a heavy user gets blocked while their colleagues' unused capacity sits idle.

Codex is priced as a feature of ChatGPT, exactly as Claude Code is priced as a feature of Claude. OpenAI's Codex pricing page states plainly that ChatGPT Work usage inside ChatGPT uses the same pricing, credits, and usage limits as Codex, and that the Pro plan comes in 5x and 20x usage variants, which is the same shape as Max. If you already pay for one of these assistants for non-coding work, the marginal cost of its coding agent is often zero, and that is the comparison a lot of buyers should actually be running.

The practical advice when comparing any two of these is to compare at your actual daily hours, not at list price. Two tools at $20 are not equivalent if one of them stops after 30 minutes. For a wider view of the category, see the best AI for coding and Windsurf pricing.

What developers actually say about the Pro plan

The most-cited critique of Pro for coding work is a Reddit post titled "Claude Pro at $20 is a trap for dev work", whose author reports hitting the cap after roughly 30 minutes of real Claude Code work on the most capable model and being locked out until the window resets.

It is worth engaging with that honestly rather than dismissing it, because the structural part of the complaint is correct and the specific number is not universal.

The structural part: Pro is positioned by Anthropic as a plan for everyday work and regular use, not for a full day of agentic coding. Nothing on the pricing page tells you in advance how many hours of agent work you get, so a developer buying Pro expecting a coding plan is buying on an assumption the vendor never made.

The part that varies: the 30-minute figure is one person, one model, one workload. In the same subreddit, other developers report getting one to two hours a day on Pro, and some report almost never hitting the limit when using Claude Code as an assistant rather than as an autonomous agent. Session cost scales with model choice, parallel agents, and how much codebase context each turn carries. That is why the same plan reads as generous to one person and unusable to another.

The practical read: Pro is a real plan for a developer who uses Claude Code in short bursts alongside other work. It is the wrong plan for anyone running multi-hour agent sessions, and the upgrade path from $20 to $100 is a fivefold jump with nothing in between.

The argument over what "5x" and "20x" mean

This dispute is now in court, so it is worth stating carefully and with the register attached: what follows is an allegation that has not been tested, not a finding.

The Verge reported on 8 September 2026 that a group of Claude subscribers filed an expanded class action alleging Anthropic deceptively advertised the limits of the Max tier. The complaint's core claim is the same structural point this post makes above: the advertised 5x and 20x apply to five-hour session windows that are themselves subject to a weekly limit, which the plaintiffs say adds up to a much smaller total increase in capacity than the marketing implies. One of the attorneys, Monica Vaca, told The Verge that understanding the terms required clicking through two hyperlinks to reach the word "session" and then visiting the Pro plan page for a definition of it.

Anthropic's current pages do document both clocks, in the Pro, Max, and Team articles cited throughout this post, and the pricing FAQ now spells out the session-and-weekly structure directly. Whether that was clear enough at the time of purchase is the question the court will answer, and nothing here predicts the outcome.

What a buyer should take from it today is narrower and not in dispute. The multiplier you are quoted is a session multiplier. Budget against the weekly ceiling, not the session one, and if the weekly ceiling is the thing that will bite you, no tier upgrade removes it; only usage credits or an API key do.

How to make a plan last

Everything in this section maps to a documented lever, so each one is checkable rather than folklore. The sources are Anthropic's cost management documentation and its models, usage, and limits in Claude Code guide (published 15 April 2026).

Clear between unrelated tasks. Every previous message in a session is resent on every turn, so a session that has wandered through three problems pays for all three on each new message. Anthropic's guide gives a usable test: if your next prompt would make sense in a brand-new terminal, run `/clear` first. Your CLAUDE.md and project files stay available; only the chat history goes. Use `/compact` instead when you are mid-task and need the summary, and note that `/clear` cannot be undone.

Match the model to the job. Anthropic's guidance is that Sonnet handles the large majority of coding work, Opus is for hard debugging, cross-cutting refactors, and architecture calls, and Haiku is for renames, log lines, and mechanical work. The built-in `/model opusplan` mode formalises the best-value pattern: plan with Opus, execute with Sonnet, without clearing the conversation in between.

Reference files by path instead of pasting them. Anything pasted sits in context in full for the rest of the session. Anthropic's guide adds a detail that surprises most people: the `@` prefix injects the entire file plus its CLAUDE.md tree, so a bare path is the token-saving version.

Keep CLAUDE.md lean and move specifics into skills. The file is loaded at session start and occupies context on every message. Anthropic's advice is to keep it under roughly 200 lines, add a note only the second time you have to correct the same mistake, and move workflow-specific instructions into skills, which load on demand rather than always.

Turn off MCP servers you are not using. MCP tool definitions are deferred by default, so only names and server instructions enter context until a tool is used, but the cost documentation still recommends running `/context` to see what is consuming space and `/mcp` to disable unused servers. It also notes that CLI tools such as `gh`, `aws`, and `gcloud` are more context-efficient than MCP servers because they add no per-tool listing at all.

Lower the effort level on routine work. Extended thinking is on by default, thinking tokens are billed as output tokens, and the default budget can run to tens of thousands of tokens per request. `/effort` lowers it. On models with a fixed thinking budget you can also set `MAX_THINKING_TOKENS`.

Be careful with agent teams. Anthropic's documentation states that agent teams use approximately seven times more tokens than standard sessions when teammates run in plan mode, because each teammate maintains its own context window as a separate Claude instance. The same warning now applies to the redesigned Claude Code Projects: The New Stack reported on 17 September 2026 that Anthropic expects Projects to hit usage limits sooner when multiple threads run, because each thread counts as a full Claude Code session.

Know the cache cliff. This is the least-known lever and one of the largest. Anthropic's cost documentation says the prompt cache lifetime is one hour on a subscription, but drops to five minutes once you are drawing on usage credits, and is five minutes by default on an API key or cloud provider. In practice, the moment you spill past your included usage, every break longer than five minutes starts reprocessing your full context at full price. If you are going to run on credits, keep sessions short and cleared.

Measure before you optimise. `/usage` shows plan usage bars, an attribution breakdown by skill, subagent, plugin and MCP server, and behaviour flags for things like long context or cache misses when one accounts for 10% or more of recent usage. `/status` shows your remaining allocation. `/insights` writes an HTML report on recent sessions. All three are cheaper than guessing.

Team and Enterprise seats

Team plans start at two members and run to 150 seats. Standard seats include 1.25x Pro's per-session allowance, Premium seats include 6.25x, and both include Claude Code. Anthropic's Team and Enterprise guide for Claude Code (published 11 June 2026) confirms Claude Code is included with every Team seat, with Premium seats carrying more usage for heavier workloads.

Usage does not pool on Team plans. Anthropic's Team article is explicit: limits are per-member, each member has their own, and one member hitting theirs does not affect anyone else. Admins mix and match seat types within one allocation, so the standard approach is to put everyone on Standard, watch one billing cycle, and promote only the confirmed heavy users.

Seat changes are asymmetric, and this costs money. Per Anthropic's seat management article, upgrading a member from Standard to Premium is prorated and charged immediately, but moving someone from Premium to Standard, or reducing your seat count, takes effect only at your next renewal with no prorated refund. Getting the mix wrong upward is cheap and reversible in days; getting it wrong downward is expensive and reversible only at renewal.

Enterprise works on a different model, and this catches teams out. Per Anthropic's Enterprise plan documentation, the seat fee covers platform access only. All usage across Claude, Claude Code, and Cowork is billed separately at standard API rates, with no included token allowance and no per-seat usage limits. The Enterprise billing article adds the operational half: self-serve Enterprise buys usage as credits up front, drawn from one shared organisation-wide balance, and when that balance hits zero usage stops for everyone until an owner buys more. Sales-assisted Enterprise is billed monthly in arrears instead, so one person's heavy week does not block colleagues, it just enlarges the invoice.

For a finance team modelling this, an Enterprise quote is a floor, not a total. The comparison against Team Premium at $100 to $125 per seat is a comparison between a fixed cost with a ceiling and a variable cost without one.

What admins actually control. On Team and seat-based Enterprise plans, owners enable usage credits and set spend limits at the organisation level, by seat tier, and per individual member, per Anthropic's Team and Enterprise usage credits article. Limits apply hierarchically: a user cannot exceed their own limit or the organisation's, whichever is lower, and hitting a personal limit pauses that member's credits until the end of the month. On the reporting side, the cost documentation describes a spend report in org analytics with per-user and per-model CSV export updated daily, an Enterprise Analytics API for programmatic per-user reporting using a key with the `read:analytics` scope, and OpenTelemetry export as the only option that streams per-user token and cost metrics into your own stack in near real time.

Two admin details that are easy to miss. First, if your organisation pays contracted rates rather than list, every cost figure Claude Code shows a developer is computed at list price unless an admin sets the `modelPricing` managed setting, which changes what is reported and not what is charged. Second, Anthropic publishes per-user token and request rate recommendations that scale down with team size, from 200k to 300k tokens per minute per user for teams of one to five, down to 10k to 15k for teams above 500. Those are the numbers to bring to a capacity conversation.

One compliance caveat worth flagging before rollout: Anthropic's Team and Enterprise guide states that on a HIPAA-ready Enterprise plan, Claude Code is included in the seat but is not covered under the HIPAA-ready offering.

Billing surprises that are structural, not bad luck

Every item here is documented behaviour rather than an edge case, which means it will happen to you if the conditions are met.

An `ANTHROPIC_API_KEY` in your environment silently overrides your subscription. Anthropic's Pro and Max guide states that if the variable is set, Claude Code authenticates with it instead of your plan, producing API charges rather than drawing on included usage. This is the single most common way a developer on a $200 plan generates an unexpected API invoice.

The prompt cache lifetime drops from one hour to five minutes the moment you spill onto usage credits. Covered above as a cost lever; it belongs here too, because the effect on the bill is invisible in the interface and arrives entirely in the form of larger charges for the same work.

Max has no annual option. Pro and Team have one, Enterprise is annual only, and the tier most heavy individual users need is monthly only. If you are budgeting a year of heavy use, there is no discount to model.

Mobile store subscriptions cannot enable usage credits. Per the usage credits article, if you subscribed through the iOS or Android app, you must enable and purchase credits on the web version, and you will need to add a payment method there first.

US-only inference costs 1.1x. Anthropic's Enterprise billing article states that organisations with the US-only inference setting turned on are billed at 1.1x standard API rates on recent models, on both self-serve and sales-assisted plans. The same multiplier appears on the pricing page for data residency.

Enterprise seats cannot be removed mid-term on self-serve plans, and seats added during a term are charged prorated immediately. Combined with the Team rule that downgrades wait for renewal, the general pattern is that expanding is instant and contracting is not.

Large invoices cannot be paid by card. Sales-assisted Enterprise invoices of $50,000 or more can only be paid by bank transfer, which is a procurement scheduling detail rather than a cost, but it derails a rollout if discovered in the last week of a quarter.

Usage credits have a daily redemption ceiling. The usage credits article puts it at $2,000 per day, and notes that in some jurisdictions, such as Japan, credits purchased from 10 September 2026 expire six months after purchase.

Parallel work multiplies consumption in ways the interface does not price. Agent teams at roughly seven times a standard session, and the new Projects beta where each thread counts as a full Claude Code session, are both cases where one instruction from you produces many billed sessions.

Usage credits, usage bundles, and the economics of overflow

Once you accept that a heavy month will exceed the plan, the interesting question is what the overflow costs and how to make it cheaper.

Usage credits are the basic mechanism: enable them in Settings, add a payment method, prepay funds, and set a monthly spend cap or leave it unlimited. Everything past your included usage bills at standard API rates. Auto-reload is available, and Anthropic's guide makes the point that credits and included usage are tracked separately in the usage dashboard, so you can see which one you are spending.

Usage bundles are the part most buyers have not noticed. Anthropic's usage bundles article (published 18 May 2026) lets you pre-purchase credit in blocks at a discount: $50 for 10% off, $250 for 20% off, and $1,000 of credit for $700, a 30% discount. Individual Pro and Max subscribers can buy up to $2,000 of discounted bundles a month, Team owners up to $3,000, and usage past those caps bills at standard rates. The balance is one pool across Claude, Claude Desktop, mobile, Claude Code, and Cowork.

For anyone who has already established that they spend a few hundred dollars a month past their plan, that 30% is the cheapest saving available and requires no change to how anyone works. The arithmetic: $1,000 of overflow bought as bundles costs $700, so a developer consistently spilling $500 a month past Max 20x saves roughly $100 to $150 a month by pre-buying rather than paying as they go. Bundles only apply after included usage is exhausted, so they are not a substitute for picking the right tier.

What changed in Claude Code pricing, and why it matters now

A tool's pricing history tells you how often your budget line is likely to move.

Primary sources: TechCrunch, 28 July 2025; Anthropic, Claude Code and new admin controls for business plans, 20 August 2025; Buy usage bundles; Manage usage credits for paid Claude plans; How am I billed for my Enterprise plan?; The New Stack, 17 September 2026.

Two patterns in that history are worth a buyer's attention.

The list prices have barely moved. Pro has been $20 since before Claude Code existed, and Max has been $100 and $200 since at least July 2025, when TechCrunch quoted both figures. What has moved, repeatedly, is the definition of what those prices buy: weekly ceilings added in 2025, Opus-specific weekly limits, seat restructuring on Enterprise, and new consumption surfaces such as agent teams and Projects that spend the same allowance faster.

That is the thing to plan around. The risk on this budget line is not a price rise. It is that your existing price buys a different amount of work than it did two quarters ago, in either direction, and that nothing prompts you to re-check.

Who should not buy a subscription

Honest answer first: several kinds of team are better served by API or cloud-provider billing, and Anthropic's own documentation effectively says so.

Anyone running Claude Code unattended. Automation, scheduled jobs, and CI runs do not have a human waiting out a five-hour reset. Anthropic's usage and limits guide describes the API-key path as having no hard stop, with the account simply charged for what it uses. If a blocked run breaks a pipeline, the cap is the problem, not the price.

Teams that need spend attributed per developer across clouds. If Claude Code runs through Amazon Bedrock, Google Cloud's Agent Platform, or Microsoft Foundry, billing goes to the cloud account and Anthropic's own analytics dashboards do not cover that usage. The documented options are OpenTelemetry export, a self-hosted Claude apps gateway, or an LLM gateway that tracks spend by key. That is a more complex setup than a subscription, and it is the only one that fits an organisation whose finance rules require cloud-committed spend.

Very heavy individual users who value not waiting over predictability. Someone consistently consuming $600 a month of equivalent usage against a $200 plan is getting excellent value until the week they get blocked on a deadline. The blunt version: if being stopped costs you more than $400 of delay in a month, the subscription is not the cheaper option.

Anyone whose real need is chat rather than code. Claude Code is included in every paid plan, so this is rarely a reason to avoid a subscription, but it does change which tier. A person who writes twenty lines of code a month and lives in chat should be on Pro regardless of how much they like the terminal.

Conversely, if your usage is human-driven, daily, and reasonably consistent, the subscription wins on arithmetic and wins again on billing sanity. The crossover sits at roughly 8 active days for Max 5x and 15 for Max 20x at Anthropic's own $13 per active day average, and most working developers clear both.

Where Coworker fits

Coworker does not compete with Claude Code, and picking one does not rule out the other. Claude Code is the coding agent. The problem Coworker solves sits next to it: giving whatever coding tool you use access to context that lives outside the repository.

Coworker MCP connects AI tools, including Claude Code, Cursor, and other MCP clients, to 50+ apps through one governed connection, so the agent can read the Jira ticket, the Slack thread, and the design doc that explain why the code is the way it is. That is a separate line item from your Claude subscription and a separate question from which Claude tier to buy. Coworker's own pricing is Pro at $29.99, Max at $149.99, and Enterprise on a custom quote.

There is a cost angle to that too, and it runs in the right direction. Anthropic's own documentation recommends disabling unused MCP servers and notes that tool definitions consume context, so consolidating a pile of individually-installed servers behind one governed connection is a context-management decision as much as a governance one. If you are evaluating which MCP servers are worth connecting in the first place, how to choose MCP servers covers the selection criteria, what MCP is covers the protocol, and enterprise MCP and MCP security cover the parts of an evaluation that never show up on a pricing page. For the wider vendor picture, see our enterprise AI pricing comparison.

Book a demo to see how Coworker MCP fits alongside your existing Claude Code setup.

Frequently asked questions

How much does Claude Code cost per month?

Claude Code is included with a Claude subscription rather than sold separately. That means $20 per month on Pro ($17 billed annually), $100 on Max 5x, $200 on Max 20x, or $25 to $125 per seat per month on Team, per Anthropic's pricing page as of 18 September 2026.

Is there a version of Claude Code that costs nothing?

No. Claude Code is not part of the Free plan. Pro at $20 per month is the entry point for subscription access, and the alternative is paying per token on the Anthropic API with your own key.

What happens when I hit my Claude Code usage limit?

Your session limit resets on a rolling five-hour window, and there is a separate weekly limit that resets at a fixed time each week. If usage credits are enabled on your account, you can keep working past the included limit and the overflow is billed at standard API rates instead of being blocked.

Is a Claude Code subscription cheaper than paying per token on the API?

For consistent daily use, usually yes. Anthropic's own cost documentation reports an average of around $13 per developer per active day at API rates, which passes the $100 Max 5x price at roughly 8 active days and the $200 Max 20x price at roughly 15. Spiky or automated workloads can still favour the API because it has no cap and no weekly quota.

What is the difference between Max 5x and Max 20x?

Price and capacity, not features. Max 5x is $100 per month and gives five times Pro's per-session allowance; Max 20x is $200 and gives twenty times. Both multipliers apply to the rolling five-hour session window, and the weekly limit is a separate constraint that does not scale by the same factor.

Do Claude Code and Claude chat share the same usage limit?

Yes. Anthropic's documentation states that usage across claude.ai, Claude Code, and Claude Desktop counts toward the same limit, and that IDE usage is metered the same way as terminal usage. A long chat session reduces what remains for the agent.

Does an Enterprise seat include Claude Code usage?

No. Anthropic's pricing page lists Enterprise at $20 per seat per month billed annually, and the seat fee covers platform access only. Usage across Claude, Claude Code, and Cowork is billed separately at standard API rates with no included token allowance, so the seat price is a floor rather than a total.

Why did I get an API bill when I am paying for a Max plan?

Check for an `ANTHROPIC_API_KEY` environment variable. Anthropic's Pro and Max guide states that if that variable is set, Claude Code uses the API key for authentication instead of your subscription, which produces API charges rather than drawing on included usage.

Can a Claude Team plan share usage across the team?

No. Anthropic's Team article states that usage limits are per-member rather than applied to the team as a whole, so one member hitting their limit does not affect anyone else's capacity, and unused capacity cannot be transferred. Pooled consumption only appears on usage-based Enterprise plans, where all usage is billed together at API rates.

What are Claude usage bundles and are they worth buying?

Usage bundles are prepaid blocks of usage credit sold at a discount: 10% off at $50, 20% at $250, and 30% at $1,000, which costs $700. They apply only after your plan's included usage is exhausted, with monthly purchase caps of $2,000 for individuals and $3,000 for Team owners. If you already overspend your plan every month, they are the cheapest available saving.

How many hours of Claude Code do I get per week?

Anthropic does not publish an hours figure on its current pages. The only official numbers came with the 2025 weekly-limit announcement, when Anthropic told TechCrunch that Pro users could expect 40 to 80 hours of Sonnet 4 per week, Max 5x 140 to 280 hours of Sonnet plus 15 to 35 of Opus, and Max 20x 240 to 480 hours of Sonnet plus 24 to 40 of Opus. Those figures are from July 2025 and name models that have been replaced, so treat them as historical shape rather than a current promise.

What is the cheapest way to cut a Claude Code bill without changing plans?

Clearing between unrelated tasks is the single largest lever, because the whole conversation is resent on every turn. After that: default to Sonnet and reserve Opus for planning, keep CLAUDE.md under roughly 200 lines, disable MCP servers you are not using, lower the effort level on routine work, and avoid agent teams for small jobs, since Anthropic puts their consumption at roughly seven times a standard session when teammates run in plan mode.

Does a Claude Code subscription work on Bedrock or Vertex?

No. Those are API-key paths billed per token to your cloud account, not subscription seats. Anthropic's documentation notes that Claude Code does not send metrics from your cloud back to Anthropic, so per-user cost attribution there needs OpenTelemetry export, a Claude apps gateway, or an LLM gateway that tracks spend by key.

Which Claude plan should a two-person startup buy?

Two Team Standard seats at $20 each on annual billing is $40 a month and adds central billing, admin controls, and the option to promote one person to a Premium seat later. Two individual Pro subscriptions cost roughly the same and carry slightly less per-session capacity with no admin tooling. If either developer already runs multi-hour agent sessions, a single Max 5x at $100 for that person plus Pro for the other is usually the better split.

Related reading

• Claude API Pricing Explained
• Windsurf Pricing Breakdown
• The Best AI for Coding
• Enterprise AI Pricing Compared
• How to Choose MCP Servers for Your Team