What Claude Code costs: $20 or $1,200 a month

The official Claude Code pricing page takes about thirty seconds to read. $20 for Pro, $100 or $200 for Max, or pay per token through the API.

It answers “what does Claude Code cost” the way a car’s sticker answers “what does it cost to own”. Accurate, and misleading. The same tool doing the same work can cost $20 a month or well over $1,200. The difference comes from choices most developers make without thinking.

Two ways to pay

Claude Code is a command-line tool that runs in your terminal and talks to Anthropic’s models. You pay for it through whatever Claude plan or API account you already have.

Subscription. It comes with every paid plan: Pro at $20 a month ($17 billed annually), Max 5x at $100 and Max 20x at $200, plus Team and Enterprise seats. It uses the same usage allowance as your Claude chats. Hit the limit and you wait for the window to reset, or pay overage at API rates. The free plan doesn’t include terminal access, so Pro is the cheapest real way in.

API. You pay per token with no monthly floor. Sonnet is $2 per million input tokens and $10 per million output (introductory pricing to 31 August 2026, $3 and $15 after). This is how people running Claude Code at scale usually pay. It’s also where the cost swings.

Why $20 or $1,200?

Anthropic’s own cost documentation says the average enterprise user spends about $13 per developer per active day. That’s $150 to $250 a month, with 90% under $30 on any active day. So a regular user lands at or above the Max plans. If you use it most days, a flat subscription beats paying per token.

Token use depends on your choices:

One developer reported using 10 billion tokens over eight months. By one third-party analysis, that would have cost more than $15,000 through the API. On a $100 Max plan it was about $800. That’s 93% less for the same work. It’s the same gap between sticker and bill as the full cost of one AI agent.

The levers

On the API, two things do most of the work.

Prompt caching charges cache reads at 10% of the normal input price. A coding session reuses the same system prompt and codebase context turn after turn. That’s exactly what caching was built for.

Batch processing takes 50% off input and output for jobs that don’t need to run in real time, like bulk analysis or overnight runs. The two discounts stack.

On a subscription, the hidden part is the limits. Usage runs against a rolling session window and a weekly cap. Pick the plan your habits fit inside without constant overage.

In short

Claude Code costs at least $20 a month. Past that, the plan price is a floor. A light user on Pro pays $20. A heavy API user running Opus without caching can pay as much as $1,200 a month by third-party estimates, for work a $200 Max plan would have covered.

The other AI coding assistants work the same way. Pick the billing path that fits your usage. Then use the levers: caching, batching, model choice and keeping context lean. For one developer, that’s a monthly annoyance. For fifty making those choices on their own, it decides whether finance can predict the bill.