DeepSeek V4 Flash is live · Our GLM-5.2 price just dropped to 50% of list
Blog
Cost control

Claude Code Pricing

Choose a subscription for a recurring allowance and hosted features, or API billing when you need metered usage, explicit model costs, and project-level spend control.

12 min readOmniaKey
Claude CodepricingsubscriptionAPI billing

Claude Code pricing has two different systems that should not be compared as if they were the same product. A Claude subscription charges a recurring price for an allowance shared across Claude surfaces. An API path charges for the model tokens Claude Code actually uses. Paid subscribers can also enable usage credits after their included allowance runs out.

The short answer is simple:

  • choose Pro or Max when you use Claude regularly, want Claude's hosted surfaces, and prefer a recurring allowance;
  • choose API billing when usage is occasional, automated, attributable to projects, or needs a hard financial boundary;
  • compare cost per accepted task, not a monthly price against one token rate.

A subscription is not a bucket of API credits. Anthropic does not publish a fixed token amount for Pro or Max. Limits depend on conversation length, model, features, and usage patterns, and can include rolling session, weekly, model, or feature limits.

Prices and product behavior were checked August 15, 2026. Public prices below are in USD before location-specific taxes. Plans, limits, model rates, and launch promotions can change; verify the linked first-party pages and the live OmniaKey catalog before purchasing.

Claude Code pricing at a glance

Anthropic's public individual pricing currently lists Free, Pro, and Max. Team and Enterprise use different seat and usage structures.

RouteCurrent public priceHow usage is measuredImportant boundary
Free$0Lower included limitsGood for evaluation, not a predictable production allowance
Pro$20 monthly, or $200 annually ($17/month equivalent)Included usage shared across Claude productsAt least 5x Free usage per five-hour session, but no fixed token promise
MaxFrom $100 monthly5x or 20x Pro usage per five-hour sessionHigher allowance, still subject to rolling and other limits
Team Standard$25/seat monthly, or $20/seat with annual billingPer-seat allowance shared across Claude productsAdmin controls and team features; not metered API credit
Team Premium$125/seat monthly, or $100/seat with annual billing5x Standard-seat usageDesigned for heavier users inside a managed team
Enterprise$20/seat/month plus usage billed at API rates, billed annuallySeat plus metered usageSecurity and administration are part of the purchase
Claude Console API, cloud, or gatewayNo Claude subscription required for the local CLITokens and provider-specific chargesHosted Claude Code surfaces and subscription-only features stay separate

The quoted plan numbers are headline prices. Taxes, annual prepayment, organization terms, and optional usage credits can change the invoice. A larger allowance is also not a guaranteed number of prompts: one short request after a long repository session can carry far more context than several clean-session questions.

What a Claude subscription pays for

Pro and Max combine Claude Code with the wider Claude product. The recurring payment covers an included usage allowance rather than a per-request invoice. That makes the monthly cash cost easy to predict until a limit is reached, but it makes token-level unit economics harder to calculate.

Subscription usage can be consumed by:

  • Claude Code in the terminal;
  • Claude chat and other included Claude surfaces;
  • long conversations and repeated context;
  • the selected model and reasoning effort;
  • subagents, agent teams, MCP tools, and other model-backed features.

The /usage screen reflects this distinction. API users see session tokens and a locally estimated cost at Anthropic's standard list rates. Subscribers see plan-usage bars, recent attribution, and behaviors that are consuming their allowance. The API dollar estimate is not relevant to the included portion of a Pro or Max bill.

When a paid subscriber reaches a plan limit, Anthropic can offer usage credits at standard API rates. This is a third billing state: the account remains a subscription account, but work beyond the included allowance becomes metered. A spending limit can cap those credits.

What API-billed Claude Code costs

API billing separates the Claude Code client from the recurring Claude plan. The local CLI can authenticate through Claude Console, a supported cloud provider, or an authorized LLM gateway. Each path has its own rate card and spend controls.

At standard Anthropic API prices, the current Claude models in the OmniaKey catalog have these per-million-token rates:

ModelAnthropic inputAnthropic cache readAnthropic outputOmniaKey launch inputOmniaKey launch cache readOmniaKey launch output
Claude Haiku 4.5$1.00$0.10$5.00$0.20$0.02$1.00
Claude Sonnet 5$2.00$0.20$10.00$0.60$0.06$3.00
Claude Opus 5$5.00$0.50$25.00$1.00$0.10$5.00
Claude Fable 5$10.00$1.00$50.00$2.00$0.20$10.00

Review the live pages for Claude Haiku 4.5, Claude Sonnet 5, Claude Opus 5, and Claude Fable 5 before choosing a model.

For Fable's dedicated official-versus-platform calculations, API access steps, and Opus cost boundary, read the Claude Fable 5 pricing guide.

OmniaKey figures are the current launch rates checked on the fact-check date, not a promise of a permanent discount. Cache-write prices, special speed modes, provider tools, taxes, retries, and other charges are excluded from this compact table. Use the live model catalog for the current gateway quote.

Claude Code API cost examples

A useful estimate starts with a token mix, not a guessed number of prompts:

text
request cost =
  uncached input × input rate
  + cached input × cache-read rate
  + cache writes × cache-write rate
  + generated and thinking tokens × output rate
  + provider tool charges

The following examples use Claude Sonnet 5, uncached input, no paid provider tools, and no retries. They illustrate scale; they do not predict how many files or turns your task will need.

Representative workloadToken mixAnthropic standard APIOmniaKey launch rate
Small, focused fix250K input + 20K output$0.70$0.21
Multi-file feature1M input + 100K output$3.00$0.90
Long investigation5M input + 500K output$15.00$4.50

The arithmetic for the multi-file example is:

text
Anthropic: (1 × $2.00) + (0.1 × $10.00) = $3.00
OmniaKey:  (1 × $0.60) + (0.1 × $3.00)  = $0.90

Real Claude Code sessions often have a large cache-read component because conversation history, project instructions, tool definitions, and previously read context may be reused. They can also produce more output-billed thinking tokens than the visible answer suggests. Measure the completed task in /usage, then reconcile it against the provider's actual usage record.

Subscription vs API: which is cheaper?

There is no reliable universal break-even point. Anthropic does not define Pro or Max as a fixed token bundle, so converting a $20 subscription into a per-million-token price creates false precision. The better decision is based on workload shape.

Your usage patternBetter starting pointWhy
Regular interactive work across Claude productsProOne recurring purchase includes Claude Code and hosted Claude surfaces
Heavy daily use with Pro limits interrupting workMax, or Pro plus controlled usage creditsMore included capacity before metering or waiting
Occasional coding sessionsAPI or prepaid gatewayPay only when requests run instead of holding a monthly seat
CI, scripts, scheduled jobs, or autonomous agentsAPI or gatewayProgrammatic credentials, per-project attribution, and explicit spend
Several coding tools and model familiesGatewayOne balance and key policy can cover compatible tools without merging their protocols
Claude Code on the Web or subscription-only featuresSubscriptionAPI credentials do not unlock hosted subscription surfaces
Team that needs identity, analytics, and administrationTeam or EnterpriseThe seat includes product and organizational controls, not only inference

For a light user, API billing can avoid paying during quiet months. For a heavy interactive user, a subscription can be more convenient and may deliver more useful work for the recurring price. For automation, a metered key is usually easier to attribute and cap even when an individual subscription exists for human work.

Using both is legitimate. Keep the subscription for interactive and hosted workflows, then give automation a separate API key with its own budget. Do not share a personal subscription login with a server or treat a consumer allowance as an automation pool.

Calculate cost per accepted task

Token price answers only part of the purchasing question. A cheap run that fails twice can cost more than an expensive run that passes once. Track a unit the team actually values:

text
accepted-task cost =
  model and tool charges
  + rerun charges
  + reviewer correction time
  + remediation cost for escaped mistakes

Use the same task class and acceptance checks when comparing routes. Record:

  • selected model and effort;
  • uncached input, cache reads, cache writes, output, and thinking;
  • number of model turns and retries;
  • whether tests and required checks passed;
  • human correction time;
  • final provider charge.

Anthropic reports that enterprise API deployments average around $13 per developer per active day and $150–250 per developer per month, with 90% remaining below $30 per active day. Those are broad deployment observations, not a forecast for an individual repository or a Pro-plan equivalent. A small pilot remains the honest way to budget.

The Claude Code token-usage guide covers context, tools, effort, and reruns in more detail. It complements this pricing decision rather than replacing it.

Control API spend without breaking legitimate work

A financial cap should be based on observed task cost. Start with one scoped key and one representative workload, then set the ceiling above normal variance.

With OmniaKey:

  1. create separate keys for local work, CI, and scheduled automation;
  2. give each key a limit that matches its responsibility;
  3. inspect requested model, tokens, latency, and actual cost per call;
  4. investigate loops or unexpected models before raising the limit;
  5. revoke one workload without interrupting the others.

The balance and usage guide owns the exact product controls. The pay-as-you-go Claude Code guide owns the legitimate authentication paths and setup boundary.

Common pricing mistakes

Comparing Pro directly with one API token rate

Pro includes multiple Claude surfaces and a variable allowance. API billing measures model consumption. Compare the work each route completes over a real period, not unlike units in one table.

Assuming Max is unlimited

Max provides more usage than Pro, not unlimited use. Rolling session, weekly, model, and feature limits can still apply.

Treating the context window as included usage

A one-million-token context window is capacity. On an API path, tokens placed in that window are still billed according to their input or cache classification.

Using /usage list-price estimates as the gateway invoice

Claude Code calculates its API dollar estimate locally at Anthropic's standard list rates. Promotions or contracted gateway rates can differ. Reconcile against the provider dashboard.

Ignoring cache writes and thinking tokens

Cache reads can be cheaper than uncached input, but cache writes have their own price. Extended thinking is billed as output on API paths even when it is not visible in the final response.

Frequently asked questions

How much does Claude Code cost per month?

Free costs $0. Pro is $20 monthly or $200 annually. Max starts at $100 monthly. API-billed Claude Code has no fixed monthly price: the bill depends on model tokens, cache behavior, tools, and retries. Taxes and organization terms can change the total.

Is Claude Code included in Claude Pro?

Yes. Claude's public pricing includes Claude Code with Pro. Its usage shares the plan's allowance with other Claude products rather than creating a separate API-token balance.

Is the Claude Code API cheaper than a subscription?

It can be for occasional or tightly scoped use because there is no recurring seat. A subscription can be better for frequent interactive use and hosted features. There is no exact token break-even because Anthropic does not publish Pro or Max as fixed token bundles.

What happens when I reach a Claude Code subscription limit?

You can wait for the displayed reset, move to a larger plan, or enable controlled usage credits on an eligible paid plan. Switching models does not reset a shared session or weekly limit, although it may help after a model-specific limit.

Can I use a subscription and an API key together?

Yes. Keep the credential paths explicit. A common setup uses subscription OAuth for interactive work and a separate capped API key for CI or automation. Verify the active path with /status and the relevant billing dashboard.

Does /usage show my real OmniaKey cost?

It shows token counts and a local estimate based on Anthropic's standard API prices. Use the OmniaKey usage dashboard for the actual gateway charge and current launch rate.

Sources