Anthropic · Command-line development agent
Claude Code: pricing, plans and quotas
What it actually costs, what the quotas mean, and when another plan becomes the rational choice.
Pricing checked on 3 August 2026
What it is
Claude Code is Anthropic's development agent. It works from a terminal, reads and edits repository files, runs commands and keeps project context between turns. It also runs inside supported IDEs, VS Code and its forks such as Cursor, plus JetBrains IDEs.
Two ways to pay, and they behave differently. Either a Claude subscription, where Claude Code usage counts against the same quota as Claude conversations. Or usage-based API billing, where you pay per million tokens with no quota ceiling. The first suits a person, the second an industrial workload.
What it costs
- Free$0Claude Code included, with standard weekly limits.
- Pro$17 per monthAnnual rate, $200 paid up front. $20 per month if billed monthly.
- Max 5x$100 per monthFive times Pro usage per session, and higher output limits.
- Max 20x$200 per monthTwenty times Pro usage per session.
- Team, standard seat$20 per seat per monthAnnual rate, $25 billed monthly. Includes Claude Code and Claude Cowork.
- Team, premium seat$100 per seat per monthAnnual rate, $125 billed monthly. Five times a standard seat's usage.
- API, Sonnet 5$2 / $10 per million tokensInput then output. Introductory rate through 31 August 2026, then $3 and $15.
- API, Opus 5$5 / $25 per million tokensInput then output. Haiku 4.5 comes down to $1 and $5.
Prices as published by the vendor, in dollars and excluding applicable tax. They change often, hence the verification date.
What the quotas mean
Anthropic publishes no message count and no hour count. Quotas are relative: Pro gives at least five times Free usage per session, Max gives five or twenty times Pro. Pages that quote you a precise number of requests per day made it up.
The mechanics are published. Limits reset on a rolling five-hour window, and paid plans add weekly limits on top. Usage is shared between Claude and Claude Code: a long chat session eats into the coding quota, and the reverse holds too.
On 6 May 2026 Anthropic doubled Claude Code's five-hour limits for Pro, Max, Team and seat-based Enterprise plans, and removed the peak-hours reduction for Pro and Max. Consumption benchmarks published before that date no longer hold.
Available models vary by account and change with releases. The /model command is the source of truth, as the documentation itself states.
What drives the bill
On the API, the line that matters is input, not output. An agent re-reads the whole session history on every turn, so a long session pays for the same files several times. That is why caching changes everything: a cache read costs ten per cent of the normal input price, a five-minute cache write costs 1.25 times, and a one-hour write twice.
Model choice spans a factor of five between Haiku and Opus. For routine coding work Anthropic's own documentation recommends Sonnet by default, and that matches what we see: switching everything to Opus is the fastest way to triple a bill with no visible gain.
Models from version 4.7 onward use a tokenizer that produces roughly thirty per cent more tokens for the same text. At an identical headline price, the real bill goes up. That kind of detail appears in no comparison table.
Batch processing returns fifty per cent on input and output, but it is asynchronous, so it is unusable for an interactive session.
Alternatives
- CodexOpenAI$20, $100 or $200 per month
Bundled with ChatGPT Plus, then Pro at $100 for five times the usage and $200 for twenty. The ladder mirrors Anthropic's almost tier for tier.
- CursorAnysphereFree, $20 or $40 per month
Free Hobby tier with limited requests, Pro at $20, and $40 per user for teams. It is an editor rather than a terminal agent, and it can drive Claude Code.
- GitHub CopilotGitHubFree, $10, $39 or $100 per month
The only one publishing actual numbers: 2,000 completions a month on Free, then monthly credits of $15, $70 and $200 by tier. Cheapest entry point.
Where we stand
This block is a field opinion, not a record. Everything above it can be checked against the vendor.
The most useful calculation is one division. A Max 20x seat costs $200 a month, exactly the price of 100 million Sonnet 5 input tokens at the introductory rate. Output, billed five times higher, and caching, ten times lower, move that marker both ways, so treat it as an order of magnitude. If your steady usage stays below it, the subscription wins. Above it, or if load arrives in spikes, the API costs less and never stops you mid-sprint.
The monthly price matters less than the cost of an interruption. A developer blocked by a quota hit at 3pm on a Tuesday costs more in lost time than the gap between two plans. That is the argument that moves most teams from Pro to Max, and it is better stated in hours than in dollars.
One billing trap keeps coming back in teams: because usage is shared between Claude and Claude Code, someone using chat heavily burns their own seat's coding quota. Nobody sees it coming, and it is hard to diagnose because the symptom is a limit reached with no matching coding work.
Which plan for which profile
- Trying it outFree then Pro
Free is enough to know whether the tool suits you. Move to Pro when you hit the limit more than once a week.
- Full-time developerMax 5x
Pro fills up in a morning on a real codebase. Max 5x is the first tier where long sessions run without being cut.
- Team of three to fifteenTeam, standard seats
$20 a seat stays below individual Pro, with central billing. Move seats to premium case by case rather than everyone at once.
- Automated workloadAPI with caching
No quota, cost measurable per request, and caching divides input by ten on the stable parts of the prompt. The only workable choice in continuous integration.
Frequently asked
- Is Claude Code free?
- Yes, Anthropic's Free plan includes Claude Code with standard weekly limits. Enough to evaluate the tool, not to work on it daily.
- How much does Claude Code cost?
- It has no price of its own: it is included in Claude subscriptions, from $0 on Free to $200 a month on Max 20x, or billed by usage through the API from $2 per million input tokens on Sonnet 5.
- How many requests per day on a subscription?
- Anthropic does not publish that number. Limits reset on a rolling five-hour window, with a weekly ceiling on paid plans, and consumption depends on session length and model. Your account's usage panel is the only reliable measure.
- Subscription or API?
- Subscription for steady human use, API for automated or spiky load. As a marker, the $200 of a Max 20x seat buys 100 million Sonnet 5 input tokens at the introductory rate, before the effect of output and caching.
Sources
The practice guide
This page is about money and limits. The guide is about use: how to hold a long session without losing the thread.
Scope your first AI agent
20 minutes to review your tools, data and the first useful case. No jargon, no commitment.