Claude Code pricing confuses a lot of people, not because it's complicated, but because Anthropic's docs tell you what each plan costs without telling you what it actually means for how you work. There are three ways to pay for Claude Code: Pro ($20/month), Max ($100/month), and direct API billing. Which one makes sense depends entirely on what you're building and how often you're building it.

At AMPL, we've run Claude Code across all three billing setups on real client work. We've hit Pro's limits mid-sprint, watched an API bill climb on a heavy automation build, and worked out exactly when the Max upgrade pays for itself. This is what we found.

Quick answer: Claude Pro costs $20/month and suits occasional or light use. Claude Max costs $100/month and gives you 5x the usage. API billing charges per token and makes sense for teams with variable usage who want cost control per project.

The Three Ways to Pay for Claude Code

Claude Code runs through Anthropic's API under the hood. But how you access that API and what you pay differs significantly across the three options. Here's what each one actually means in practice.

Claude Pro ($20/month) — what you get and where it limits you

Pro is Anthropic's standard subscription. You get access to Claude's most capable models plus Claude Code for $20/month. That sounds like a reasonable entry point, and for some use cases it is.

The catch is usage limits. Pro doesn't give you unlimited access. It operates on a rolling window, typically resetting every few hours. Anthropic doesn't publish the exact numbers because they vary with server load, but in practice Pro gives you roughly 10-20 messages per hour with the heavier models before you hit a slowdown or a wait period.

For browsing, drafting, and casual use, Pro is fine. For Claude Code specifically, where a single agentic task might involve dozens of back-and-forth tool calls, file reads, and code edits, Pro runs out fast. We've hit the limit in under 30 minutes on a reasonably complex build task. It's not that the plan is bad. It's just not designed for sustained agentic coding sessions.

Claude Max ($100/month) — who this is actually for

Max is $100/month and gives you 5x the usage of Pro. There's also a $200/month tier that gives you 20x. For most people running Claude Code seriously, the $100 tier is the relevant one.

The honest answer to whether Max is worth it: if you're using Claude Code more than a couple of hours a day, yes. The maths is straightforward. At $20/month you're paying $20 for limited access. At $100/month you're paying 5x more for 5x the access. If Claude Code is genuinely part of how you build things, running full tasks, editing across multiple files, working through longer projects, the limits on Pro become friction you'll feel constantly.

We upgraded to Max on client builds after the third time we hit a limit mid-task and had to wait 20 minutes. That wait cost more in lost time than the price difference between the plans. If your time is worth anything, Max pays for itself quickly.

API billing — when usage-based makes more sense

API billing is different in kind, not just in degree. Instead of a monthly subscription, you pay per token. Input and output are priced separately, and costs vary by model. Claude Opus 4 costs $15 per million input tokens and $75 per million output tokens at time of writing. Sonnet 4 is considerably cheaper.

This makes sense in a few specific situations. First, if you're building systems that use Claude Code programmatically as part of an automated pipeline, you need API access anyway. Second, if your usage is lumpy, quiet weeks followed by intensive sprints, API billing can work out cheaper than paying $100/month for Max when you're not using it. Third, if you need to track cost per client project, API billing gives you that visibility in a way a flat subscription doesn't.

The downside is that API bills can surprise you. A heavy automation sprint at AMPL ran up a four-figure API bill in a week. Not a problem if you've budgeted for it. A problem if you haven't.

What 'Usage Limits' Actually Mean in Practice

Anthropic is deliberately vague about the exact numbers for Pro and Max limits. Here's what that actually looks like when you're working.

How fast you burn through Pro limits on real tasks

A Claude Code session on a non-trivial task, say building a new feature that involves reading several files, writing new code, running tests, and iterating, can involve 30 to 50 or more tool calls. Each one counts against your limit. A moderately complex task might consume a significant portion of your hourly Pro allowance in one go.

In practice, we found Pro workable for tasks that take 10 to 15 minutes of Claude Code work. Anything longer, or anything where you're doing multiple tasks in a session, and you start bumping into the limit. The limit doesn't cut you off hard. It slows you down and eventually asks you to wait. But that friction adds up across a working day.

When Max pays for itself

Max pays for itself when you're using Claude Code for two or more hours of actual work per day. At that usage level, the $80/month difference between Pro and Max is covered by time you're not waiting for limits to reset.

It also pays for itself on longer tasks. Claude Code handles multi-hour agentic sessions well when you have Max. You can set it on a complex task and let it run. On Pro, you'd need to break that into chunks and manage the limits manually. That management overhead has a real cost.

Cost Scenarios: What Real Usage Costs

Three realistic usage profiles and what each one actually costs.

Solo developer running light projects

Someone using Claude Code for personal projects, learning, or occasional client work, a few hours a week on lighter tasks. Pro at $20/month is probably fine here. You'll hit limits occasionally but not constantly, and the $80/month saving over Max is meaningful if you're not getting full value from the higher tier.

If you find yourself regularly hitting limits and stopping to wait, that's the signal to upgrade. Don't upgrade pre-emptively. Upgrade when the friction becomes real.

Agency or consultant running multiple client builds

This is where Pro breaks down fast. Running Claude Code across multiple active client projects means context-switching between builds, running multiple tasks in a day, and often pushing through longer sessions to hit delivery deadlines. We run Max at AMPL and still use API billing for some project-specific work where we need cost transparency per client.

If you're billing clients for time, the cost of waiting for Pro limits to reset shows up directly in your day. Max at $100/month is the minimum viable setup for anyone doing this seriously. For heavier agency work, API billing on top of or instead of Max can make more sense depending on how predictable your monthly volume is.

Business owner using Claude Code for operations

A business owner using Claude Code to build internal tools, automate processes, or work through operational systems isn't necessarily coding for hours a day. They might be running intense sessions once or twice a week, then light usage the rest of the time.

For this profile, Max is worth it if those intensive sessions are regular. If you're genuinely running Claude Code hard for a full day once a week, the limits on Pro will frustrate you and slow down what you're trying to build. Max removes that friction. If it's more occasional than that, Pro might be enough, or API billing if you want to pay for exactly what you use.

Hidden Costs to Budget For

A few things that don't show up in the headline pricing.

API cost on top of subscription. If you're using Claude Code via the API rather than a subscription plan, costs scale with complexity. Long context windows, feeding Claude a large codebase to work with, use a lot of tokens. A single task on a large project can cost $5 to $15 in API calls. Fine if you know it's coming. Budget for it.

Context window economics. Claude's context window is large, which is one of the reasons it's so useful for coding. But larger context means more tokens, which means higher API costs. If you're on API billing, how you structure prompts and how much context you load affects your bill meaningfully.

The cost of interruption. This one's not on the invoice but it's real. Every time you hit a Pro limit mid-task and wait 20 minutes, that's 20 minutes of lost momentum. If you're doing serious work with Claude Code, the opportunity cost of the cheaper plan often exceeds the price difference with Max.

Model choice. On API billing, the model you use changes the cost significantly. Opus 4 is the most capable and the most expensive. Sonnet 4 is considerably cheaper and handles most coding tasks well. For lighter tasks, defaulting to Sonnet and saving Opus for the hard problems is a sensible approach.

Our Recommendation by Use Case

To make this concrete.

Start with Pro if you're new to Claude Code, doing occasional work, or genuinely unsure how much you'll use it. See where the limits bite. Don't spend $100/month before you know you need it.

Upgrade to Max if you're hitting Pro limits regularly, running Claude Code for more than an hour or two most working days, or doing project work where stopping mid-task costs you real time. The $80/month difference pays for itself quickly if you're using Claude Code seriously.

Use API billing if you're building systems that call Claude programmatically, you need per-project cost visibility, your usage is highly variable, or you're running intensive sprints where paying for actual usage makes more sense than a flat monthly fee. Pair this with careful model selection.

Combine Max and API if you're an agency or consultant doing both interactive Claude Code work and building automated pipelines. The subscription covers your interactive work, API covers the programmatic stuff, and you get clear cost separation.

One thing worth saying clearly: don't let plan cost be the reason you underuse Claude Code on a real project. The time savings on a well-structured build are significant. If you're doing serious work with it, the plan cost is a rounding error compared to the value.

If you're trying to work out what Claude Code could actually do for your business operations specifically, that's a conversation worth having. Book a free audit at amplconsulting.ai and we can map out what the real ROI looks like for your setup.