Which Claude Code plan you actually need: Pro, Max, or API (after using them all)

TL;DR The Free plan doesn't include Claude Code (you need Pro at minimum, or pay per token via the API). For daily coding, Pro runs out fast: I live between Max 5x ($100) and Max 20x ($200), moving between them based on how much inference I'm about to burn. But the plan barely matters: if you don't reach for the biggest model for everything, go easy on MCP servers, and compact often, Max is plenty. This is my experience with all of them, not a price table. (Prices as of July 2026, they change: verify at claude.com/pricing.)

Almost everything written about "Claude Code plans" is a price table copied from the official site. Not this. I've been through every individual plan (and Team), and what I want to tell you isn't what each one costs, it's which one you actually need and why.

First, the myth to kill: the Free plan does not include Claude Code. The official docs say it flat out: "The free Claude.ai plan does not include Claude Code access." You need Pro at minimum, or pay per token with an API account. So "is it free?" has a short answer: no.

The word that drives this whole decision is inference. Every time Claude replies, that's an inference: it processes your context and generates a response, and that burns tokens counting against your plan's limits (the 5-hour window and weekly cap). Picking a plan isn't picking a price, it's estimating how much inference you burn per week. And how much you burn depends less on the plan than on how you use the tool.

The plans, from experience

Free. Doesn't include Claude Code. Full stop. There's no way to code for free day to day: you need Pro at minimum, or pay per token with an API account.

Pro ($20/mo). The entry point. For me, Pro doesn't cut it for day-to-day coding: the limits run dry by mid-morning once you push it. Good for trying it seriously and for light use, not for living inside Claude Code.

Max 5x ($100) and Max 20x ($200). This is where real coding happens. I move between the two based on the inference I expect: normal weeks, 5x; heavy weeks (lots of agents, big refactors, long sessions), 20x. The "5x" and "20x" are 5x and 20x Pro's limits.

Team, Enterprise, and API. I can speak to Team from experience (per-seat, the pool is shared with chat and Cowork). On Enterprise and API/Console I'm not the right person: I have little experience there, so I won't sell you smoke. Honestly: they're per-token or per-seat, for organizations, and the docs, for API-billed deployments, put the real cost at around $150-250 per month per developer. That's where it crosses over with a subscription: if your metered spend would land around there, a flat Max caps your cost; if you're light or spiky, metered API can win.

The plan barely matters: use the tool well

This is the part almost nobody tells you. You can live on Max without touching the limits if you do four things:

• Don't reach for the biggest model for everything. Using Fable (or whatever the largest model is) for every little thing is a rookie move: it's the most expensive inference for tasks that don't need it. Opus and Sonnet, used well, go a very long way. Pick the model for the task, not the most powerful one by default.
• Go easy on MCP servers. Each MCP server puts definitions into your context, and context is tokens on every turn. Connect the ones you actually use. The 10 habits to spend less are exactly about this.
• Compact often. /compact between tasks and /clear when you switch topics. A session that keeps growing makes every turn pricier (even with caching). Watch what survives a /compact.
• Know the basics: prompting, adherence, context. If you handle those and the harness (tools, skills), staying within the limits is easy. It's not about paying more, it's about spending better.

My unfiltered take: whoever complains that "Claude Code runs out too fast" is almost always using Opus for everything, with fifteen MCP servers loaded and three-hour sessions with no compaction. Fix that before you upgrade your plan.

Reference (July 2026, prices change)

Anthropic moves prices and limits every few months. Before you pay, verify at claude.com/pricing and check your own bars in /usage.

Official docs: Manage costs effectively

Requirements

A Pro, Max, Team, or Enterprise plan, or an API account with credits. The Free plan won't do for Claude Code.