Claude Code Pricing in 2026: Every Plan, Real Limits, and What It Actually Costs

Claude Code starts at $20 a month on an individual subscription, offers Max from $100 a month, and can also run against metered API credits. The new monthly API credits on Max and Team do not cover interactive Claude Code or extra usage; your rolling five-hour allowance and weekly cap still matter. Plans and rates were verified on October 8, 2026.

For interactive coding, most solo builders want Pro at $20/month, and the moment Claude Code starts stopping you mid-task most days, you move to Max 5x at $100. If you also build apps and agents on the API, the new credits change that part of your budget. The plans below were re-verified against Anthropic's live pricing page, Claude Code's cost documentation, and Anthropic's help, model, and changelog pages on October 8, 2026.

A quick note on what "Claude Code" even is here, because it changes how you read the price. Claude Code is Anthropic's coding agent that runs in your terminal or IDE, the thing you hand a task and it edits files, runs commands, and iterates. There is no separate flat-rate Claude Code subscription. The current paid Claude lineup includes it alongside the chat app, while API use is metered separately. That bundling is the single most misread fact about its pricing, and it drives most of the questions below.

Is Claude Code free, and is it still on the $20 plan?

No. Claude's Free plan costs $0, but Anthropic's current pricing FAQ says Claude Code is included in all current paid plans. The individual floor is Pro at $20/month or $200/year. Free gives you the Claude apps, not Claude Code under a subscription.

And yes, Claude Code is still on the $20 Pro plan. The official support article is titled "Use Claude Code with your Pro or Max plan," and it spells out that one unified subscription gives you Claude on web, desktop, and mobile and Claude Code in your terminal. Pro includes it.

Every plan, what it costs, what you get

Here is the full individual lineup with the real billing detail, because "annual equivalent" pricing hides where the money actually goes.

Pro is $20 billed monthly, or $200 billed up front for the year, which works out to about $17/month if you commit to twelve months. For that you get standard usage capacity across Claude and Claude Code. For a founder validating an idea or an indie hacker shipping a small product, Pro is the first plan to try. Exact capacity varies with conversation length, model, effort, and features, so Anthropic does not promise a fixed number of coding hours. It is the floor for Claude Code subscription access.

Max 5x is $100/month, monthly billing only, and gives you 5x the per-session capacity of Pro. This is the tier you move to when Pro interrupts you too often, the one for someone coding most of the day or running the agent on larger repositories where a single task chews through more context.

Max 20x is $200/month and gives you 20x Pro's per-session capacity, with monthly billing only. Anthropic's Max plan support page confirms both tier prices; these are web subscription prices, and mobile pricing can vary. This is for people who run Claude Code hard all day, often several agent sessions at once, where even Max 5x runs dry before the day does.

Claude Code defaults to Opus 5.5 at medium effort on Pro, Max 5x, Max 20x, Team Standard, Team Premium, Enterprise, and direct Anthropic API accounts. That account-type default applies from Claude Code 2.1.280; your saved model and effort choices, environment variables, or organization settings can change it. Effort controls how much reasoning the model does. For the cheaper everyday model, /model sonnet selects Sonnet 5.5 on a direct Anthropic connection in Claude Code 2.1.284 or later, also at medium effort by default.

Haiku 5.5 is available in Claude Code on a direct Anthropic connection from version 2.1.293. The model configuration docs map /model haiku to Haiku 5.5 there. The subagent docs also allow haiku or a full model ID in a custom subagent's model field; use claude-haiku-5-5 to pin the version. Provider aliases and organization settings still apply, so a third-party provider's haiku alias can point to Haiku 4.5.

The jump from Pro to Max 5x is 5x the price for 5x the per-session capacity, so on paper it is linear. In practice you upgrade not because you do five times the work, but because Pro keeps interrupting you and the lost flow is worth $80 more a month. That is the real trigger, and it is a usage question, not a math question, which is why the limits matter more than the sticker.

The limits that actually decide your bill

This is the part the price tables skip, and it is the part that determines whether $20 feels generous or stingy. Inside a subscription's included usage, Claude Code does not bill you per task. You buy an allowance, and the question is how fast you spend it.

Four things shape that allowance:

It is shared with the chat app. Included subscription usage in Claude on the web and Claude Code draws from the same pool. If you spend your morning brainstorming in the chat app, you have less left for Claude Code that afternoon. People who treat the two as separate budgets get surprised.

Paid subscription usage resets on a rolling five-hour session window, with weekly limits on top. Pro, Max, Team, and legacy seat-based Enterprise users can see both windows under Settings > Usage or in Claude Code with /usage. You can be inside the current five-hour window and still be throttled because the week's allowance is spent.

Fable access now depends on your tier. Max, Team Premium, and legacy seat-based Enterprise Premium include Fable use up to 50% of the weekly limit, not 50% extra. Pro, Team Standard, and legacy seat-based Enterprise Standard use pay-as-you-go usage credits for Fable from the first token. Current usage-based Enterprise plans meter every model at API rates instead of supplying a seat allowance.

Heavy agent runs are expensive in allowance terms. When you point Claude Code at a big task and let it loop, reading files, running commands, re-reading output, it consumes far more than a chat back-and-forth. That is why "why is it so expensive" usually means "my agent runs are eating the week," not a hidden charge.

When you do hit the limit, you are not stuck. Here is how to keep working, and how to keep it from costing you by accident.

1. 1

   Decide whether to keep going or wait

   At the limit, Claude Code can continue with usage credits billed at standard API rates, or you can wait for your allowance to reset. On Claude Code 2.1.234 or later, /rate-limit-options can wait and resume the interrupted task automatically. For a non-urgent task, waiting is free. For a sprint, paid usage credits keep you moving. The new monthly Max and Team API credits cannot pay for this extra usage.

2. 2

   Stay strictly on your plan

   To avoid surprise charges, keep usage credits disabled or decline them when prompted, log in with only your subscription credentials, and make sure ANTHROPIC_API_KEY is not set in that shell. Then Claude Code can only spend your included allowance.

3. 3

   Watch your usage with /usage

   Run /usage inside Claude Code to see your current plan bars, activity statistics, and usage breakdown before you start a big agent run. It is the cheapest way to avoid getting cut off mid-task.

If you find yourself hitting the wall most days on Pro, that is the signal to move to Max 5x. If you are hitting it on Max 5x, Max 20x or pay-as-you-go is the next step, covered below.

Subscription vs API pay-as-you-go

The subscription is not the only way to run Claude Code. You can point it at the Anthropic API and pay per token, with no monthly API subscription required. Whether that is cheaper depends entirely on how much you run. Interactive Claude Code remains separately metered even when your Console organization has a monthly API credit grant.

Since October 7, 2026, Max and Team include monthly API credits: $100 on Max 5x, $200 on Max 20x, and $20 per Team Standard seat or $100 per Premium seat, pooled and capped at $500 per month. The same announcement launched Haiku 5.5 and halved Sonnet 5.5 cache-read prices. The credits reduce the cost of apps and agents you build; they add no interactive coding allowance.

The credits cover the Claude API, including Message Batches, the Claude Console Playground, Managed Agents, and the Claude Agent SDK. They do not cover interactive Claude Code in the terminal, IDE, desktop, or web, or extra usage in Claude, Claude Code, or Cowork. Free, Pro, and Enterprise are not eligible for this monthly grant.

You can claim after seven days on an active eligible plan. A Max subscriber uses Settings > Billing; a Team Owner or Primary Owner uses Organization settings > Billing. In the API credits section, choose Link organization and connect a Console organization where you hold an Owner, Admin, or Billing role. Credits arrive each billing cycle, monthly on annual plans, expire at that cycle's end, and are spent before purchased credits. Everyone with an API key in the linked organization draws from that balance.

For non-interactive work, self-run claude -p and Agent SDK calls can use the grant when authenticated with an API key from that linked organization. Signing in with your Claude plan instead uses the plan's limits. Runs started by the Claude Code GitHub Action, an IDE extension, or the desktop app are excluded even with -p, as the credit rules specify.

These are the current API rates, re-verified against Anthropic's live rate card on October 8, 2026. The latest lineup is Fable 5.1, Opus 5.5, Sonnet 5.5, and Haiku 5.5. Haiku 4.5 remains available as a legacy model. MTok means one million tokens.

Since September 28, 2026, Claude Sonnet 5.5 is the latest Sonnet release. Its $2 fresh-input and $10 output rates per million tokens match the previous model, so those rates stay the same when you move everyday coding to 5.5. Sonnet 5.5 cache reads are now cheaper than Sonnet 5's.

Legacy: Sonnet 5 is now listed under Legacy models at $2 input and $10 output per million tokens.

Since September 19, 2026, Claude Code 2.1.278 has asked for eligible auto-mode classifier checks server-side by default for Enterprise and Claude API accounts, with platform-dependent rollout on Bedrock, Google Cloud's Agent Platform, Microsoft Foundry, and gateways. A server-side check carries no classifier charge. /status shows Auto mode server: Enabled; if those checks cannot reach the session, Claude Code warns before falling back to its own classifier requests, which are billed as before. This removes classifier overhead from eligible metered sessions without changing any subscription or token price.

Since September 1, 2026, Claude Code 2.1.257 has made Fable 5.1 the default Fable model on a direct Anthropic connection. Fresh input and output stay at $10 and $50 per million tokens, but cache reads fell from $1 on Fable 5 to $0.25 on Fable 5.1: ten reads of a one-million-token cached prefix fall from $10 to $2.50. The cheaper loop comes with stricter API rules, because forced tool selection now returns HTTP 400 and replayed thinking blocks can fail after earlier history is edited; the Fable 5.1 explainer has the migration sequence.

Two levers cut the API bill hard. Prompt caching lets you reuse a large context such as a codebase or long system prompt. Current cache-read rates per million tokens are Fable 5.1 at $0.25, Opus 5.5 at $0.20, Sonnet 5.5 at $0.10, Haiku 5.5 at $0.01 for prompts up to 100k tokens or $0.05 above that threshold, and Haiku 4.5 at $0.10. Ten reads of a one-million-token Sonnet 5.5 cached prefix now cost $1 instead of $2, excluding the original input, cache write, and output charges. And batch processing saves 50% on input and output for workloads that do not need an instant answer.

So when does API beat a subscription? For interactive Claude Code, it wins when your metered total stays below the flat fee. For your own apps and agents, API access is the route, and eligible Max or Team credits can offset that spend. At Fable 5.1's $50 output rate, two million output tokens alone equal $100 and four million equal $200, before input. Caching shifts that crossover only for repeated input. A Max subscription makes steady daily use more predictable, but it still has five-hour and weekly limits; the API remains a meter, with eligible app and agent usage drawing from your claimed grant before purchased credits.

If you already pay $100/month for Max 5x and also spend $100/month on eligible API work, the claimed $100 grant can reduce that purchased API spend to $0 within the billing cycle. Your Max fee stays $100, and interactive Claude Code extra usage still costs separately. That is the budget benefit: fund the app alongside your coding plan, without treating the credit as a refund or more coding capacity.

Teams and enterprise

For groups, pricing moves to per-seat, and the important change is that both Team seat types now include Claude Code. Premium buys more capacity, not access to the coding agent.

Team Standard is $20/seat/month on annual billing, or $25/seat billed monthly, for teams of 2 to 150 people. It includes Claude Code and carries 1.25x Pro usage per session. Team Premium is $100/seat/month annual, or $125 monthly, also includes Claude Code, and carries 6.25x Pro usage, which is 5x Standard. Included Claude and Claude Code usage limits are per member, not pooled across the team, and owners can enable usage credits at standard API rates after a member exhausts the included allowance. You can mix seat types on one plan, so the economical setup is Standard for most people and Premium only for the members who repeatedly exhaust Standard. The separate monthly API credits are pooled: three Standard seats and two Premium seats receive $260 ($20 x 3 + $100 x 2), subject to the $500 organization cap. That API pool cannot top up a member's interactive Claude Code allowance.

New Enterprise plans use one Enterprise seat at $20/seat/month, billed annually, plus usage at standard API rates. The seat fee covers access to Claude, Claude Code, and Cowork but includes no token allowance; current usage-based Enterprise has no plan-level or seat-level usage cap because every token is metered. Self-serve starts at 20 seats and sales-assisted at 50. Older Chat, Chat + Claude Code, and Standard or Premium Enterprise seats remain legacy tiers until the contract renews. Enterprise adds controls such as SCIM, audit logs, custom data retention, compliance APIs, and organization or user spend limits.

Which plan is right for you

The decision rule is simple: start on Pro, and let your interruptions tell you when to upgrade. You do not need to predict your usage, you need to notice when Claude Code stops you mid-task more than occasionally. That is the only signal that matters.

For most people reading this, the honest answer is Pro for a month, then watch your /usage. If you are curious how the cost stacks up against the obvious alternatives, the math against Cursor is laid out in Claude Code vs Cursor, and if budget is the hard constraint, Gemini CLI vs Claude Code covers the cheaper route.

Is the $20 Claude Code plan worth it?

For most solo builders, yes. Pro at $20/month includes both Claude Code and the Claude apps, and it is the cheapest subscription with Code access. Capacity varies by task and model, so upgrade only if Claude Code starts cutting you off mid-task most days, at which point Max 5x is the move.

How do I run Claude Code for free, or as cheaply as possible?

Claude Code is not included on the $0 Free subscription. The cheapest subscription route is Pro at $20/month, or about $17/month if you pay annually ($200 up front). API access has no subscription fee but bills every token, and there is no standing Max discount beyond occasional official promotions.

Is Claude Code Max worth it?

If you hit Pro's limits most days, Max 5x at $100/month is the value tier, 5x the per-session capacity for 5x the price. Max 20x at $200/month only makes sense if you run the agent hard all day or several sessions at once and still run dry on Max 5x. Weekly limits still apply to both. Their $100 and $200 monthly API grants fund eligible apps and agents, not interactive Claude Code or extra usage.

Was Claude Code removed from the $20 plan?

No. Claude Code is included in the Pro plan ($20/month) as well as Max. Anthropic's current pricing and support documentation explicitly cover using Claude Code on Pro or Max.

Why is Claude Code so expensive?

It usually is not a hidden charge, it is your allowance running out. You are paying for frontier-model tokens, and a heavy agent run reads files, runs commands, and loops, which burns far more of your shared and weekly allowance than a chat conversation. Watching /usage and using a Max tier for sustained work is how you stop hitting the wall.