Claude plans in 2026: which one do you need?

The question behind the price list

Nobody actually searches for "Claude pricing" because they want to read a price table. They search because they're stuck on a decision: is Claude Code included in the Pro plan or is that a separate purchase? Should a developer pay monthly or per token? Does a five-person team buy five Pro accounts or a Team plan? I've answered some version of these in support tickets for a year now, usually from customers setting up Claude Code on one of our servers, so this is the decision guide I wish I could just link.

The lineup, as of July 2026: Free, Pro, two Max tiers, Team (with two seat types), and Enterprise — plus the API, which is not a plan at all but a separate pay-per-token account that most individual users never need and most developers eventually do. The single most-asked question has a one-line answer: Claude Code is included in every paid plan, from Pro up, and is not in the free tier. Everything after that is about matching your usage shape to a tier.

Prices move. I'll quote the numbers that were on claude.com/pricing the day I wrote this, with dates attached — treat that page as the source of truth, not this post (and not the dozens of third-party "Claude pricing" posts that are already stale).

What actually gates the tiers

Strip away the marketing and four things separate the plans:

• Usage limits. This is the real product. Every tier above Free is priced as a multiple of Pro's allowance — Max is literally sold as "5x" and "20x" Pro usage per session. Under the hood these limits are token accounting, which is why a heavy Claude Code session burns through them faster than chat ever will — Claude Code's token usage is dominated by re-sent context, not by what you type.
• Claude Code inclusion. Paid plans yes, Free no. Pro, Max, Team (both seat types), and Enterprise all cover Claude Code in the terminal and in supported IDEs — VS Code, Cursor and other forks, JetBrains — against the same shared limit as chat.
• Feature and model access. Pro and up get the full model picker and unlimited projects; Max adds higher output limits and priority access to new models during peak traffic. The free tier is not a crippled toy, though — as of July 2026 it includes web search, file creation, code execution, connectors, and extended thinking.
• Admin surface. SSO, central billing, and admin controls arrive at Team; SCIM, audit logs, the Compliance API, custom data retention, and customer-managed encryption keys arrive at Enterprise. If those acronyms mean nothing to you, you are not the Enterprise customer, and that's fine.

Notice what's not on the list: intelligence. A Free user and a Max user largely talk to the same models. You are buying volume and plumbing, not a smarter Claude.

The casual user: stay Free until it hurts, then Pro

If you use Claude a few times a week — drafting, explaining, the occasional script — Free is genuinely enough, and I'd tell you to stay on it. The limits are tight but they reset, and the feature set covers most conversational use.

Two triggers should push you to Pro. The first is hitting the usage wall mid-task more than once a week; being cut off in the middle of something you're paying attention to is worth $20 to stop. The second is wanting Claude Code at all, since Free simply doesn't have it. As of July 2026 Pro runs $20/month billed monthly or $17/month on annual billing — the annual discount is real money if you know you'll stay, but I'd do one monthly cycle first to confirm the limits fit you.

What Pro does not fix: it will not make limits disappear. Pro's allowance is shared across Claude chat and Claude Code, so the day you start using Claude Code seriously is the day Pro starts feeling small. That's not a defect, it's the tier boundary doing its job.

The developer: subscription or API?

This is the fork that generates the most confusion, because both paths run the same models and the pricing shapes are opposites.

A subscription is a flat fee with a ceiling. You pay $20 (Pro) or $100/$200 (Max 5x / Max 20x, as of July 2026) and you get a usage allowance that resets on a rolling basis, with weekly caps layered on top on some plans — the support docs describe the current windows. When you hit the ceiling, work stops until reset, unless you explicitly opt in to paid usage credits. Anthropic is clear that spilling over into per-token billing never happens without your consent.

The API is a meter with no ceiling. You pay per million tokens, in and out. As of July 2026: Haiku 4.5 at $1 in / $5 out, Sonnet 5 at an introductory $2 / $10 through August 31, 2026 (then $3 / $15), Opus 4.8 at $5 / $25, and Fable 5 — the premium tier — at $10 / $50. Batch processing halves all of it, and prompt-cache reads cost roughly a tenth of the input price. Check claude.com/pricing before budgeting; the Sonnet intro price alone will have changed by the time many of you read this.

The decision rule I give customers is about who initiates the request:

1. A human at a keyboard, working interactively — you, in Claude Code, all day. Subscription. Interactive coding is exactly the bursty, high-volume, unpredictable usage that flat pricing is designed to absorb. A day of serious agentic coding can pass through tens of millions of tokens of context re-sends; at API rates that's a scary invoice, on Max it's Tuesday. Start on Pro; if you hit the wall most days, Max 5x; if you run parallel sessions or long autonomous tasks, Max 20x. This holds whether Claude Code runs on your laptop or on a server — plenty of people keep Claude Code running on a VPS inside tmux precisely so long sessions survive a dropped SSH connection, and it bills against the same subscription either way.
2. A program, initiating requests on its own — a cron job, a webhook handler, a chatbot, anything server-side. API, no exceptions. Subscriptions are consumer accounts with consumer terms; automation belongs on an API key with per-token billing that scales from pocket-change experiments to production traffic. If that's where you're headed, start with your first Claude API app on a VPS — the account, the key, and the first working script.
3. Both — the common end state. A Max plan for your own daily coding and an API key for the things you build. They're separate accounts with separate billing, and that separation is a feature.

One failure mode catches people constantly, and it's worth stating exactly: if ANTHROPIC_API_KEY is set in your environment, Claude Code authenticates with that key and bills your API account at per-token rates — your Pro or Max subscription is silently ignored. You find out when the Console invoice arrives. Before assuming Claude Code is using your subscription:

echo "${ANTHROPIC_API_KEY:+key is set - Claude Code will bill the API}"
unset ANTHROPIC_API_KEY   # and remove it from ~/.bashrc or ~/.profile

Inside Claude Code, /status shows which account you're authenticated against. Check it once on every new machine; on a VPS especially, an old key exported in a dotfile from a previous project is the classic culprit.

The team: five people is the threshold

Team exists for groups of 5 to 150 and, as of July 2026, comes in two seat types: a standard seat at $20/month annual ($25 monthly) with about 1.25x Pro's per-session usage, and a premium seat at $100/month annual ($125 monthly) with 6.25x — the premium seat is the one meant for people living in Claude Code, and both seat types include it. Seats can be mixed, which is the point: two developers on premium, three writers on standard, one invoice.

The arithmetic against individual plans is unremarkable on purpose — a standard seat costs about what Pro costs. What you're actually buying is the plumbing: central billing instead of five expense reports, SSO, admin controls over who has access, and a no-training-on-your-data default that individual consumer accounts don't contractually promise your employer. The moment "can we get the company to pay for this?" is asked, Team is the answer — not because the seats are cheaper, but because the alternative is shared logins and someone's personal card, both of which end badly.

Watch the seat-type fit, though. A developer who maxes out a standard seat's weekly cap every week is a premium seat wearing the wrong badge; the upgrade is cheaper than the stalled afternoons. Premium seats carry two weekly limits — one across all models and one for Sonnet-class models specifically — so heavy users should skim the Team plan docs before assuming the multiplier is the whole story.

The enterprise: a different billing model, not just a bigger plan

Enterprise changed shape and most of the third-party pricing posts haven't caught up. As of July 2026 it is not "contact sales for a mystery number": there's a self-serve tier with a 20-seat minimum you can buy directly, and a sales-assisted tier from 50 seats for custom contracts. The pricing model inverts the consumer plans — a modest per-seat fee (about $20/seat/month, billed annually) that covers access only, with all usage billed separately at API rates. No per-seat usage ceilings at all: your heaviest Claude Code user never hits a wall, and your bill scales with actual consumption instead of seat-count guesswork.

That inversion is the tell for who should buy it. If your organization's problem is predictability — a fixed line item per person — Team is your plan. If the problem is capacity — power users blowing through premium-seat limits — or compliance — audit logs, SCIM provisioning, the Compliance API, custom data retention, customer-managed encryption keys, US-only inference, a HIPAA-ready configuration — Enterprise is the only tier that has any of it. Nobody buys Enterprise for the chat experience; they buy it because security review said no to everything else.

The whole decision in six lines

• Occasional chat, no coding: Free, until the limits annoy you.
• Regular use, or you want Claude Code at all: Pro.
• Claude Code most of the day, hitting Pro's wall: Max 5x, then 20x.
• Anything programmatic — bots, pipelines, server-side calls: API, regardless of what subscription you also hold.
• 5+ people, company money, admin needs: Team, mixing standard and premium seats.
• Compliance requirements or uncapped power users: Enterprise.

And whichever branch you land on, verify the two numbers that actually matter on claude.com/pricing the day you buy: the price of your tier and the shape of its usage limit. Both have changed in the last twelve months. Both will change again.

FAQ

Is Claude Code included in the Pro plan?

Yes. As of July 2026, the $20/month Pro plan includes Claude Code in the terminal and in supported IDEs (VS Code and its forks, JetBrains), billed against the same usage allowance as Claude chat — there is no separate Claude Code subscription. The caveat: an exported ANTHROPIC_API_KEY environment variable makes Claude Code bill your API account instead, so unset it if you want your subscription to cover the usage.

Does the free tier include Claude Code?

No. As of July 2026 the free plan covers Claude on web, desktop, and mobile — including web search, file creation, and code execution — but not Claude Code. To use Claude Code you need either a paid plan (Pro and up) or an API key with pay-per-token billing; the cheapest way in for regular use is Pro.

Should I get a Claude subscription or use the API?

Decide by who initiates the requests. A human working interactively — especially in Claude Code — belongs on a subscription, because flat pricing absorbs the huge token volumes of agentic coding. A program making requests on its own (cron jobs, chatbots, pipelines) belongs on the API, where per-token billing scales with load and there's no consumer usage ceiling. Most developers eventually hold both, and that's the intended shape.

What do the higher tiers actually get you?

Volume and plumbing, not smarter models. Max buys multiples of Pro's usage allowance (5x or 20x per session as of July 2026), higher output limits, and priority access during peak traffic. Team adds the organizational layer — SSO, admin controls, central billing, mixable standard and premium seats. Enterprise switches to seat-plus-usage billing with no per-seat limits and adds the compliance surface: SCIM, audit logs, custom data retention, and a HIPAA-ready option.