Skip to content AI News: Mistral Large 4 is open weight: what you can run, and what it costs on the API →

Claude Code Pricing: Every Plan, the Usage Limits and the API Math

Claude Code has no price of its own. It comes with every paid Claude plan, starting with Pro at $20 a month, or $17 a month billed yearly, and the Free plan doesn't include it. You can also skip the subscription and pay per token through a Claude Console account or a cloud provider. On a subscription, your terminal sessions and your chats draw from one shared set of usage limits.

Every way to pay for Claude Code

There are two kinds: a flat subscription with usage limits, and a per-token bill whose only ceiling is the one you set. Prices are in US dollars before tax.

OptionPriceClaude Code usageFits
Free$0Not includedChat only
Pro$20 a month, or $17 a month billed yearly ($200 up front)Included; at least 5x the chat usage Free gets per five-hour sessionA first month, to learn where your limit falls
Max 5x$100 a month, billed monthly5x ProCoding every day
Max 20x$200 a month20x ProYou hit Max 5x limits
Team, Standard seat$25 a seat monthly, $20 billed yearly, for teams of 2 to 150Included, more than ProTeams that want one bill and SSO
Team, Premium seat$125 a seat monthly, $100 billed yearly5x a Standard seatThe heavy coders on a team
Enterprise$20 a seat a month billed yearly, 20-seat minimum, plus usage at API ratesIncluded; usage-based plans have no per-seat limitCompanies that need SCIM provisioning
Claude Console (API)Per token; Sonnet 5.5 is $2 in and $10 out per millionNo plan limit; it stops at the spend limit you set on the workspaceCI pipelines and the odd session
Amazon Bedrock, Google Cloud's Agent Platform, Microsoft FoundryPer tokenBilled, and capped with budget controls, in that cloud's own consoleAWS, Google Cloud or Azure shops

Max comes with $100 or $200 a month in API credits and Team with up to $500 a month in pooled API credits, yet those credits can't be spent in Claude Code or the Claude apps. Model access also differs: Pro reaches Fable only through usage credits, while Max gets it for up to 50% of its weekly limits.

Start on Pro. When the limit keeps stopping work you need to finish, Anthropic's own advice to Pro users who consistently hit limits is Max 5x. A team of two or more pays $3 a seat more than annual Pro for Team Standard and gets central billing plus a bigger allowance. Bursty or unattended work belongs on an API key. If you're still picking a tool, set these prices against Cursor's plans and its per-token Claude billing or which ChatGPT plans include Codex.

How the usage limits work

Each paid plan meters you twice: a session limit on a rolling five-hour window, plus a weekly limit. Anthropic publishes no message or token counts, only the multiples in the table above.

The weekly limit resets at a fixed time assigned to your account, shown in Settings > Usage on claude.ai, and Max adds a separate weekly limit for Fable.

If the session limit lands mid-response, the reply may run on to finish its current step, showing Usage limit reached · wrapping up. Pro gets that once per weekly period; Max can get it at every session limit. Anthropic says the wrap-up is rolling out gradually, so your account may not have it yet.

Expect Claude Code to burn through a window faster than chat. Anthropic's admin guidance says one debugging session can consume more than a day of chat, because every turn carries file contents and tool calls.

What happens when you hit the limit

Claude Code stops and tells you when the window resets, unless usage credits are on. Then your requests switch to standard API rates and keep going until you reach your monthly spend limit or the prepaid balance runs out.

Two charges arrive without any limit being hit. Fast mode costs $8 in and $40 out per million tokens on Opus 5.5, and on a subscription it's paid from usage credits even when your plan has usage left. And if an ANTHROPIC_API_KEY variable is set on your machine, Claude Code bills that key and leaves your plan untouched. Check with:

echo $ANTHROPIC_API_KEY

On Pro with a deadline, turn on usage credits with a small monthly cap. If you'd never accept an extra charge, keep them off and log in with only your plan credentials.

What a Claude Code session costs on the API

On an API key you pay list price per million tokens, and cache reads are the cheapest line, which matters because Claude Code caches the conversation it resends each turn. Per million tokens, the models Claude Code runs on today cost this on Anthropic's API price list:

ModelInput5-min cache writeCache readOutput
Fable 5.1$10$12.50$0.25$50
Opus 5.5$4$5$0.20$20
Sonnet 5.5$2$2.50$0.10$10
Haiku 5.5, prompts up to 100K tokens$0.10$0.125$0.01$0.50
Haiku 5.5, longer prompts$0.50$0.625$0.05$2.50

One-hour cache writes cost more ($8 on Opus 5.5, $4 on Sonnet 5.5). On an API key or a cloud provider, Claude Code defaults to the five-minute cache, so the five-minute column is the one you'll pay.

The costs doc prints a sample /usage readout: a Sonnet 4.6 session with 1.2k input, 5.3k output, 940k cache-read and 50k cache-write tokens, totalling $0.55. Sonnet 4.6's legacy rates reproduce it exactly:

Sonnet 4.6 ($3 in, $15 out, $0.30 cache read, $3.75 cache write)
  input           1,200 x $3.00  / 1M = $0.0036
  output          5,300 x $15.00 / 1M = $0.0795
  cache read    940,000 x $0.30  / 1M = $0.2820
  cache write    50,000 x $3.75  / 1M = $0.1875
  total                                 $0.5526  ($0.55)

Same tokens on Sonnet 5.5
  0.0024 + 0.0530 + 0.0940 + 0.1250   = $0.2744  ($0.27)

Same tokens on Opus 5.5
  0.0048 + 0.1060 + 0.1880 + 0.2500   = $0.5488  ($0.55)

Sonnet 5.5 with nothing from the cache
  991,200 x $2.00 / 1M + $0.0530      = $2.0354  ($2.04)

The cache cuts that session to about a seventh of its uncached price. It's also why your first message after a long break costs more: past the cache lifetime, Claude Code reprocesses your full context.

For a monthly figure, Anthropic reports averages across enterprise deployments of about $13 per developer per active day and $150 to $250 per developer per month, with 90% of users under $30 a day. At $13 a day, Pro's $20 equals about one and a half active days and Max 5x's $100 nearly eight. If you code with Claude Code more than a couple of days a month and the plan's limits hold, the subscription costs less.

How to check what you're spending

Run /usage inside Claude Code; /cost and /stats are aliases for it. What it shows depends on how you pay.

For scripted runs, cap the spend per call. --max-budget-usd stops a print-mode run once its estimated spend reaches the amount, though spend can overshoot it:

claude -p --max-budget-usd 5.00 "Run the test suite and fix the first failing test"

On an API key, set a workspace spend limit before your first long session. On a subscription, open /usage daily for a week and see which skill or MCP server is eating the window.

Ways to spend less

Anthropic's costs doc traces unexpectedly high API spend to two habits: sessions never cleared and Opus left as the default model.

  1. Clear between tasks. /clear costs nothing. /compact is itself a large request because it reads the whole conversation it summarizes. Run /rename first if you'll want the session back.
  2. Default to Sonnet. Save Opus for architecture and multi-step reasoning, and give simple subagents model: haiku. At list price, Opus 5.5 is twice Sonnet 5.5 on every line.
  3. Lower the effort. Thinking tokens bill as output, so a lower /effort on routine work cuts the bill.
  4. Trim MCP servers. /context shows what fills the window and /mcp disables servers you aren't using. CLI tools such as gh add no per-tool listing at all.
  5. Filter output with hooks. A hook can cut a 10,000-line log down to the matching lines before Claude reads it, and you can copy working hook configs to start from.
  6. Move workflows into skills. CLAUDE.md loads on every request and a skill only when used. Anthropic suggests keeping CLAUDE.md under 200 lines, holding only the project facts Claude needs every session.
  7. Watch agent teams. With teammates in plan mode they use about 7x the tokens of a standard session.
  8. Stop early. Press Escape the moment Claude goes off course.

A vague request like "improve this codebase" sends Claude scanning the whole repo. Name the files, the test and the stopping point:

Fix the failing test tests/test_invoice.py::test_rounding.
Read only src/billing/invoice.py and that test file; don't open anything else unless the error points there.
Run only this one test. When it passes, stop and show me the diff.

Sources

Weekly

New prompts in your inbox

One email a week: the best new prompts and one short guide. Confirm by email and get the free Prompt Cheat Sheet (PDF). Unsubscribe any time. Privacy policy