Claude Code Pricing: Complete Guide and UK Costs

Is Claude Code Free?

Technically, no. While you can install the NPM package for free, you cannot authenticate or use the tool without a paid workspace.

Claude Code Pricing Explained

Using Claude Code incurs costs in two ways: Token Usage (intelligence) and Compute (execution).

1. Token Usage Costs

Claude Code runs on whichever current Claude model you point it at, typically Opus 5.5 for hard work and Claude Sonnet 5 for everything else. Every time it reads your files, plans a task, or writes code, it consumes tokens.

• Pro/Max Users: Usage draws from your daily message limits. Claude Code is very hungry, a single complex task can consume 50-100 message equivalents.
• API Users: You pay per token. An average coding session costs $1–$5 (~£0.80–£4) depending on complexity.

2. Compute Costs (Container Runtime)

When Claude lists files or runs terminal commands, it uses a secure sandboxed container.

Usage Limits: The Real Constraint

For subscription users, the number on the invoice is rarely the number that matters. What actually governs your day is the usage limit, and Anthropic meters every paid plan two ways at once.

• A session limit that resets every five hours. A session starts with your first message and runs for five hours; usage accumulates against that window's ceiling. You can start a new session afterwards, and several can occur in a week, but each has its own cap.
• A weekly limit across all models. This resets at a fixed time assigned to your account rather than on your billing date, which catches people out. It is the cap that heavy Claude Code users hit first.

The difference between Pro, Max 5x and Max 20x is essentially how much capacity sits behind those two caps. Max 5x buys roughly five times Pro's per-session capacity for $100/month (approximately £80), and Max 20x roughly twenty times for $200/month (approximately £160). Nothing about the tool changes; you simply get further into a working day before it stops.

Two other consumption factors matter more than most people expect. First, model choice: an Opus-class model burns through a given allowance substantially faster than a Sonnet-class one for the same work, on the order of 1.7 times by some measurements. Second, parallelism: running several subagents at once multiplies token consumption by roughly the number of agents. A five-agent fan-out is a genuinely useful technique and also the quickest way to exhaust a weekly cap.

If you need to keep working past the included limit, most plans let you enable usage credits that continue at standard API rates rather than simply cutting you off. A Console (API) account, by contrast, has no five-hour sessions and no weekly caps at all, which is why heavy automation and CI use tends to end up there.

API Pricing Per Model

If you authenticate with an API key instead of a subscription, you leave the world of caps and enter the world of per-token billing. These are the published Anthropic rates per million tokens, with indicative sterling conversions.

Note the shape of those numbers: output costs five times input on every tier. Claude Code reads far more than it writes, so input volume drives the bill on exploration-heavy work while output drives it on generation-heavy work. It also means extended reasoning is expensive, because thinking tokens bill at the output rate.

How to Keep Costs Down

Four levers do most of the work, and they apply whether you are managing a token bill or a weekly cap.

1. Use prompt caching. A long, stable system prompt, a CLAUDE.md file and a fixed tool schema get resent on every turn. Caching cuts the cost of that repeated input dramatically, and agentic sessions are exactly the workload it was designed for.
2. Batch anything asynchronous. Work that does not need an answer within seconds, bulk evaluation, classification, large-scale documentation passes, qualifies for the Batch API at half the standard rate.
3. Right-size the model per task. Reading files and running searches does not need frontier reasoning. Pinning exploration subagents to a cheaper model while reserving Opus for the actual patch is the single highest-leverage change most teams can make.
4. Keep context tight. Every token in the window is re-read on every turn. Clearing a session between unrelated tasks, and keeping CLAUDE.md concise, costs nothing and saves more than most micro-optimisations.

One configuration detail causes more surprise invoices than anything else: if ANTHROPIC_API_KEY is set in your environment, Claude Code will bill through the API rather than your subscription. If you have both a Max plan and an API key on the same machine, check which one you are actually spending.

How to Download Claude Code

Claude Code is distributed as an NPM package. You need Node.js installed on your machine.

Installation Steps

Claude Code vs Cursor vs Copilot

How does Anthropic's terminal agent compare to the leading AI code editors?

The headline prices flatter Copilot and mislead slightly on both of the others. Cursor's entry tier is $20/month (approximately £16), with Pro+ at $60 (approximately £48), Ultra at $200 (approximately £160) and Teams at $40 per user (approximately £32). GitHub Copilot starts at $10/month (approximately £8) for Pro, rising to $39 for Pro+, with Business at $19 per user and Enterprise at $39 per user. On 01/06/2026 Copilot moved away from premium request units to usage-based GitHub AI Credits, where one credit equals $0.01 and you consume credits against each model's published per-million-token rate, effectively the same economics as a raw API bill.

That shift matters for the comparison. All three products have converged on the same underlying structure: a modest subscription that includes a bounded allowance, followed by metered consumption at something close to API cost. The differences that remain are about interface and autonomy rather than price. Claude Code lives in the terminal and will happily run for an hour unattended; Cursor is an editor you drive; Copilot is an assistant that completes what you are already typing.

The practical consequence is that Claude Code consumes far more tokens per task than either competitor, because reading a codebase, planning, editing and re-running tests is simply more work than autocompleting a function. Comparing the $20 subscriptions like for like is therefore misleading: you are comparing an assistant that spends tokens when you type with an agent that spends them continuously while it works.

Which Plan Should You Choose?

There is no universally correct answer, but the decision resolves quickly once you know how you work.

• Occasional use, a few tasks a week: Pro at $20/month (approximately £16) is enough. If you are hitting the cap once a fortnight, it is still cheaper to wait than to upgrade.
• Daily driver, several hours a day: Max 5x at $100/month (approximately £80). This is the tier most full-time developers using Claude Code seriously end up on, and the point at which running out mid-afternoon stops being routine.
• Parallel agents, all-day autonomous work: Max 20x at $200/month (approximately £160). If you are regularly fanning out subagents across a large codebase, the lower tiers will not keep up.
• Teams: note that standard Team seats do not include Claude Code. Team Premium at $125/seat/month (approximately £100, or $100/seat on annual billing) is the tier that does, so budget per developer rather than per organisation.
• CI, automation and bursty workloads: use an API key. No session windows, no weekly caps, and you only pay for what you run. Add prompt caching and the Batch API and the effective rate drops further.

A reasonable rule of thumb: if your monthly API spend on Claude Code would exceed the price of a Max plan, switch to the subscription; if it would not, stay on the API and keep the flexibility. For the wider plan comparison beyond Claude Code, see our Claude AI pricing guide, and for the tooling side of getting value out of the agent, our guide to Claude Code workflows.

Free Guide

Get the free guide: Claude vs ChatGPT, Gemini & Grok

A 20-page playbook covering everything you need to choose and use the big four AI models in 2026, full cost and feature comparisons, what each is best (and worst) at, and how-tos for images, vectors, building a website, Claude Code and more.

Pop your email in to get it free

Download free guide Email me the guide and occasional AI tool insights. No spam, unsubscribe anytime. See our privacy policy.