Claude Code Pricing: Every Plan, Token Cost, and What We Actually Spend

The first time I saw the Claude Code invoice for our team, I thought something was broken. Four developers, one week of real usage, and the number felt low. So I spent an afternoon pulling every pricing page, every help center article, and every support thread I could find, then I mapped it against what we were actually seeing in our own Anthropic Console. This is the result.

If you just want the short version: most individual developers should start on the Pro plan at $20 a month and only upgrade once they hit a wall. Teams should almost always mix Standard and Premium seats. And if anyone tells you Claude Code "charges per token" without asking what plan you're on, they are about to confuse you. Let me explain why.

#The six ways to pay for Claude Code

Claude Code is one product with six distinct billing paths. This confuses almost everyone, so here is the short map before we go deep on each one.

1. Pro at $20 a month. Included usage, good for a few focused sessions a day.
2. Max 5x at $100 a month. Five times the Pro allowance.
3. Max 20x at $200 a month. Twenty times the Pro allowance, plus priority during peak hours.
4. Team Standard at $20 per seat per month (annual) or $25 monthly. Five seat minimum. Note: this does not include Claude Code.
5. Team Premium at $100 per seat per month (annual) or $125 monthly. This is how teams get Claude Code with centralized billing.
6. Direct API via the Anthropic API with pay per token billing. No subscription, you are charged for exactly what you use.

The trick is that the subscription plans (Pro, Max, Team Premium) hide the token math behind an included usage allowance, while the API plan makes you pay for every token. Both route to the same underlying Claude models. The question is which one saves you money given how you actually work.

#Pro plan: $20 a month

This is where almost every solo developer should start. For twenty dollars, you get Claude Code in the terminal, VS Code extension, JetBrains plugin, the desktop app, and web. You get access to both Claude Sonnet 4.6 and Claude Opus 4.6. And you get an included usage allowance that covers what Anthropic describes as "focused coding sessions."

The catch is the rolling five-hour window. Pro allows roughly 45 messages every five hours, which in practice is about three to four real coding sessions a day if you are being efficient. If you run Claude Code like a background worker on multiple tasks at once, you will hit that limit before lunch.

I ran Pro for about two weeks on a real workload. It held up for a single developer doing focused feature work. It fell apart the moment I tried to use it as a multi-agent system running in parallel. Your mileage will vary based on how "agentic" your workflow is.

#Max 5x and Max 20x: $100 and $200 a month

The Max tiers exist because power users kept hitting the Pro ceiling. Max 5x gives you five times the Pro allowance, which translates to roughly 225 messages per five-hour window. Max 20x gives you twenty times, so around 900 messages per five-hour window plus priority access during peak load.

What does that buy you in real terms? I pulled together numbers from the official Claude help center, Anthropic's cost management docs, and a couple of independent analyses to build a rough hours-per-week picture.

If you are running Claude Code for a full workday with Opus as your primary model, Max 20x is the only plan that will stay out of your way. The $200 price tag sounds expensive until you work out the API-equivalent cost.

Here is the math I ran for my own usage. I averaged about 50,000 input tokens and 15,000 output tokens per real Claude Code session, running maybe eight sessions a day. At Opus API rates, that is roughly $0.625 per session, or $5 a day. Sounds cheap. But once I added the running cost of keeping the codebase indexed, the cost of multi-agent runs, and a few occasional Fast Mode bursts (which are six times the base rate), my real API-equivalent number would have been north of $9 a day. Over a month, that is $270 on API or $200 on Max 20x with fewer surprises.

#The Team plans: Standard and Premium

Here is where every article I read got this wrong at least once. The Team plan has two seat types and mixing them is the whole point.

• Team Standard at $20 per seat per month (annual), $25 monthly. Five seat minimum. Includes Claude, not Claude Code.
• Team Premium at $100 per seat per month (annual), $125 monthly. Includes Claude Code and Cowork, plus five times the usage of Standard.

The mix and match structure is the lever. If you have a team of twelve with four developers, five product managers, and three support folks, you do not need everyone on Premium. You put the four developers on Premium ($400 a month), the other eight on Standard ($160 a month), and your total is $560 a month for centralized billing, admin controls, and Claude Code for exactly the people who need it.

Where this gets interesting is the weekly rate limit. Teams share limits at the workspace level, which means Anthropic expects that not every developer uses Claude Code at the same moment. If you have a small team where everyone is hands-on-keyboard all day, budget for higher per-seat spend. If you have a large team where coding happens in bursts, the shared pool works in your favor.

#API direct: pay per token

If you would rather skip the subscription entirely, you can point Claude Code at a raw API key and pay for exactly what you use. Here are the verified rates from Anthropic's pricing docs, current as of April 2026.

A few things stand out when you look at these numbers in context.

Output tokens are five times more expensive than input tokens on every model. That means your cost is dominated by how much Claude writes back, not how much of your codebase you send in. A "help me read this 200KB file" question is cheap. A "rewrite this 200KB file" is not.

Sonnet 4.6 is 60% cheaper than Opus 4.6 for every token category. And according to Anthropic's own cost management docs, Sonnet handles most coding tasks well. Their recommendation is to reserve Opus for architectural decisions and multi-step reasoning, which matches what I found running both for a month.

Cache reads are a tenth of the input price. This is a big deal for how Claude Code actually works. Claude Code uses prompt caching aggressively when you are in a session, which means the second, third, and fourth messages in a conversation are dramatically cheaper than the first. If you exit and restart often, you throw that caching away.

#What we actually spend

This is the part everyone wants to know. Our team runs a mix of Max 20x and Premium team seats, and I have been tracking daily cost via the /cost command and our Anthropic Console dashboards. Here is what the real distribution looks like.

These numbers line up almost exactly with what Anthropic reports publicly. Their March 2026 data says the average developer spends around $6 a day with 90% of users staying under $12 a day. Their enterprise cost management page quotes a higher $13 a day per active developer with 90% under $30. The difference is workload intensity, not a contradiction.

My takeaway is this. If you are a solo developer doing focused work, you will likely spend $3 to $7 a day if you were paying API rates. That is well under the $20 Pro plan, so Pro is effectively unlimited for you. If you are a power user running parallel agents and leaning on Opus, you are looking at $10 to $15 a day, which is right at the Max 5x break-even. If you run Claude Code as a full-day background worker and use Opus heavily, you will push past the Max 20x break-even only if you are extremely active.

#Plan comparison: which one wins for you

The honest recommendation hierarchy for individual developers is this. Start on Pro. If you hit the cap more than twice a week, move to Max 5x. If you hit the cap on Max 5x, move to Max 20x or reconsider your workflow. If you bounce off Max 20x, you are probably running Claude Code in a way where the API direct plan makes more sense because you want hard spending limits rather than hitting a wall.

For teams of five or more developers who want centralized billing, Team Premium is almost always the right answer. Below five seats or when you need hard spending ceilings instead of included allowances, direct API through the Anthropic Console wins.

#Five things I wish I had known before signing up

I made most of these mistakes so you do not have to.

Prompt caching pays off faster than you think. Anthropic's pricing page lays out the math: a five-minute cache costs 1.25x the input price to write and 0.1x to read. That means caching pays off after a single cache read. If your workflow restarts conversations often, you are leaving real money on the table.

Fast Mode is a trap if you forget to turn it off. Fast Mode on Opus 4.6 multiplies your bill by six. It is great for urgent one-off work. It is a disaster as a default.

The /cost command is your friend. It shows your current-session token usage. If you are on API direct, this is the single most useful thing in the Claude Code CLI. Pair it with the /stats command if you are on a subscription.

Sonnet 4.6 is underrated. I ran a week of A/B work between Sonnet 4.6 and Opus 4.6 on my own codebase. Sonnet handled 85% of the tasks I threw at it, and I could not tell the difference on output quality for anything that was not an architecture decision. At 60% the cost, Sonnet is the default I recommend now. If you want the long version of this argument, I wrote it up in our Claude Code six month review.

The five-hour window resets on a rolling basis, not at midnight. I spent a stressful afternoon thinking I had bricked my account when I hit the Pro cap at 2 PM. The window rolls, so your limit will free up throughout the day. You do not have to wait until tomorrow.

#FAQ

#Related reading

• Claude Code: The Complete 2026 Guide for the full feature overview
• Is Claude Code Free? for the honest answer on free paths
• Claude Code vs Cursor if you are also comparing editors
• Claude Code vs OpenAI Codex
• Claude Code Review: Six Months

#Sources