Claude Code pricing trips people up because it doesn't work the way most software subscriptions do. You're not buying a separate product. You're choosing how to access the same underlying model, and the right answer depends almost entirely on how you actually use it.

At AMPL, we've run Claude Code across Pro, Max, and API billing at different points depending on the project. This isn't a walkthrough of Anthropic's pricing page. It's a breakdown of where each tier makes sense and where it starts costing you more than it should.

How Claude Code billing actually works (it's not a separate product)

Claude Code is a terminal-based coding tool that runs on top of Anthropic's models. It's not a standalone subscription. It runs through whichever Anthropic plan or API access you already have.

What you're paying for, in every case, is token usage. Every message you send, every file Claude reads, every response it generates. That's all tokens. The plan you're on determines how many tokens you get and what happens when you hit the limit.

This matters because Claude Code is unusually token-heavy. It reads context, writes code, checks its own output, iterates. A single solid session on a complex file can burn through more tokens than a day of regular Claude chat usage. So the plan that works fine for everyday Claude use might not survive an afternoon of serious development work.

The four pricing paths: Free, Pro, Max, and API

There are four ways to access Claude Code. They share the same underlying model but have very different cost structures and usage limits.

Free tier — what you get and where it breaks down

The free tier gives you enough to try Claude Code and get a feel for what it can do. That's basically it.

Rate limits kick in fast during real development sessions. You'll hit a wall mid-task, wait for the limit to reset, and lose the thread of what you were building. For anything more than a quick experiment, that friction makes it hard to actually evaluate Claude Code properly.

Use the free tier to decide whether Claude Code is worth exploring. Not to do real work.

Pro ($20/month) — who this is actually for

Pro gives you roughly five times the usage of the free tier. For a lot of developers, that sounds like plenty. In practice, it depends heavily on your workflow.

If you're using Claude Code occasionally, a few sessions a week, specific tasks rather than full-day development, Pro is probably enough. You'll get meaningful value without hitting limits constantly.

The problem is that Claude Code encourages longer sessions. You start with a quick fix, Claude suggests a refactor, you explore it, you go deeper. Before long you've burned through a large chunk of your daily allowance on work that didn't feel that intensive.

Pro works well for solo developers who are selective about when they reach for Claude Code. It breaks down if you want it open and available throughout a working day.

Max ($100/$200/month) — when the jump is worth it

Max exists because a lot of Pro users were hitting limits and finding it disruptive. The $100 tier gives you roughly 5x Pro usage, and the $200 tier roughly doubles that again.

To be honest, the jump from Pro to Max feels steep. You're going from $20 to $100, a 5x increase in cost. But if you're regularly hitting Pro limits, the disruption to your workflow probably costs you more than the $80 difference.

The $100 tier suits developers who use Claude Code as a primary tool most days. The $200 tier is for heavy users, people doing extended sessions, working across large codebases, or using Claude Code as a core part of a team workflow.

One thing Max doesn't solve: it's still a monthly subscription with usage caps. If your usage is unpredictable, quiet some weeks and very heavy others, you might find yourself on the wrong tier half the time. That's where API billing starts making more sense.

API billing — the right choice for automation and power users

API billing is different in structure. You're not paying a flat monthly fee. You're paying per token, metered as you go. Current pricing for Claude Sonnet 3.5 is $3 per million input tokens and $15 per million output tokens (these change, so check Anthropic's pricing page for current rates).

For individual developers doing conversational coding sessions, API billing often works out more expensive than Max unless you're very high-volume. The per-token cost adds up faster than a flat subscription.

Where API billing becomes clearly the right choice:

• Automated pipelines, running Claude Code as part of a CI/CD process, code review automation, or any programmatic workflow

• Variable usage, if some months you barely touch it and other months you're using it constantly, paying per token is fairer

• Teams, multiple people or systems sharing a single billing account, where a flat subscription per person doesn't make sense

• Production integrations, building something that calls Claude Code as a service

At AMPL, we default to API billing for any automation work. The predictability of per-token cost makes budgeting for client projects much cleaner than estimating against a subscription cap.

Real usage cost breakdown by workflow type

The plan comparison makes more sense when you map it to actual usage patterns. Here's how we'd think about it for three common situations.

Solo developer doing occasional refactors

You're not a full-time developer, or Claude Code is one of several tools you use rather than the main one. Maybe you're using it to clean up code a few times a week, fix specific bugs, or get a second opinion on architecture decisions.

Pro at $20/month is probably the right call. You won't consistently hit the limits, and you'll get solid value relative to what you're paying. If you find yourself hitting walls regularly over a month or two, that's the signal to move up.

Building and iterating on a full project daily

Claude Code is open most of your working day. You're building features, debugging, refactoring, and using Claude Code as a thought partner throughout. Sessions run long and you're working across multiple files in a codebase.

This is where Pro breaks down. You'll hit limits in the afternoon on heavy days, which is exactly when you don't want the friction. Max at $100 is the more honest starting point here. If you're doing this six or seven days a week and your sessions regularly run several hours, the $200 tier might be worth it. But test at $100 first and see where you land.

Running Claude Code in automated pipelines

You're not a human sitting in a terminal session, or at least not always. Claude Code is part of an automated process: reviewing PRs, generating tests, running against code on a schedule, or integrated into a product you're building.

API billing is the right structure for this. Subscriptions are designed for individual human usage. They don't map cleanly to programmatic calls that might run constantly for a week and then barely at all. Pay per token, monitor your spend, and budget per project or pipeline rather than per month.

One honest caveat: if you're new to API billing and your automated pipeline is poorly optimised, sending large unnecessary context on every call for example, costs can climb faster than expected. Build in some monitoring before you let it run unsupervised.

How to decide which plan fits your situation

The honest version of this decision is pretty simple. Ask yourself three things:

How often do you actually use Claude Code? A few times a week casually is Pro territory. Most of your working day is Max territory. Programmatic or automated is API territory.

Is your usage predictable? If it varies a lot month to month, API billing's pay-as-you-go structure will serve you better than a fixed subscription that fits half the time.

Is this for development work or automation? Development sessions where a human is actively working suit subscriptions. Automated pipelines and integrations suit API billing almost always.

A few things that trip people up. First, don't over-optimise at the start. Start at Pro, use it properly for a few weeks, and see whether you hit limits. That's more useful data than trying to estimate your token consumption in advance.

Second, the jump to Max is large. Make sure you're actually hitting Pro limits regularly before you commit. One heavy week doesn't mean you need Max.

Third, if you're building something with Claude Code as a component, not just using it as a developer tool, API billing is almost certainly the right structure regardless of usage volume. The flexibility matters more than the per-token cost.

FAQ

Is Claude Code a separate subscription from regular Claude?

No. Claude Code runs through your existing Anthropic plan or API access. You're not buying a new product. You're choosing how to access the model. If you're already on Pro, you can use Claude Code within your existing Pro allowance. If you want Claude Code for API-based automation, you set up API access separately.

How does Claude Code Pro compare to Claude Code Max for daily development?

Pro works for occasional development sessions, a few times a week, specific tasks. For daily development where Claude Code is open most of the day, Pro limits will disrupt your workflow. Max at $100/month gives you roughly 5x the usage, which is enough for most full-time users. The $200 Max tier is for very heavy or extended daily sessions.

Can I use Claude Code API without a subscription?

Yes. API access is separate from the subscription tiers. You set up an Anthropic API account, add billing, and pay per token used. You don't need a Pro or Max subscription to access Claude Code via API. This is the right approach for automated pipelines or if you want pay-as-you-go billing rather than a flat monthly fee.

What happens when you hit Claude Code's usage limits?

On subscription tiers, Claude Code will slow down or pause when you hit limits. You might be able to continue at a reduced rate, or you'll need to wait for the limit to reset. On API billing there are no hard usage caps. You pay for what you use, so there's no cutoff. That's one reason API billing suits uninterrupted automated workflows better than subscriptions.

Is Claude Code worth the cost compared to GitHub Copilot or Cursor?

Different tools, honestly. Copilot and Cursor are IDE integrations focused on autocomplete and in-editor suggestions. Claude Code is a terminal-based agent that can read your codebase, execute commands, write and edit multiple files, and reason through complex tasks. They're not direct substitutes. For agentic coding tasks where you want Claude to work through a problem with genuine autonomy, Claude Code is in a different category. For quick inline suggestions while you write, an IDE integration probably fits better.

How do I estimate my Claude Code API costs before committing?

Anthropic's API pricing is per million tokens. A rough guide: a typical conversational message is around 200-500 tokens, but Claude Code sessions are heavier because they include file context. A solid hour-long session working through a real codebase might consume 50,000-200,000 tokens depending on file sizes and how much back-and-forth happens. Start with a test session via API, check your token usage in the Anthropic console, and scale from there. Don't commit to production usage without monitoring a few representative sessions first.

If this is something you're trying to work out for a specific project or automation build, it's usually worth spending an hour mapping it out properly before committing to a billing structure. If you'd like a second opinion on the setup, we're happy to take a look. Book a free audit at amplconsulting.ai.