"How much does Claude Code actually cost per month?" — answered. Three scenarios (personal, corporate trial, production) with monthly estimates
18,100 people search this every month, yet nobody writes "here's what it costs for your case." We break down the 5 plans (Pro, Max, Team, Enterprise, API) across three scenarios — personal trial, corporate PoC, production rollout — with monthly cost estimates.
“Claude Code — how much per month?”
I’ve gotten this question from more than 10 people in the past month. They all share one thing: they can’t find the answer.
Open the official page and the price sheet is right there. Pro $20, Max $100, Max $200, Team, Enterprise, API. The numbers are clear enough. But which plan fits your usage, and what does that come out to per month? That part doesn’t read off the table. So everyone gets stuck at “I’ll just start free.”
I was the same way at first. I read the price sheet three times, still couldn’t decide, and ended up leaving Anthropic’s pricing page open in a browser tab for two weeks. What got me moving was breaking my own usage into “three scenarios.”
⚠️ Pricing changes. This article reflects research as of May 2026. Always confirm current numbers on the official Anthropic pricing page.
Why “How much is Claude Code per month?” can’t be answered in one line
Let me be honest upfront. Answering this question in one line is nearly impossible.
There are three reasons.
First, Claude Code has two billing tracks: flat-rate and usage-based. Pro, Max, Team, and Enterprise are flat monthly subscriptions. API usage via Anthropic Console is metered per token. Same Claude Code, but the moment you walk through a different door, the invoice logic changes completely.
Second, even within the same plan, actual cost shifts with how you use it. Subscriptions have “5-hour rolling usage windows” and “weekly total caps” (Anthropic official). At 8 hours/week of hobby use vs. 160 hours/month of professional use, the same Max 5x $100 feels like two different products.
Third, the JPY conversion moves daily. As of May 2026 the rate sits around ¥150/USD, with analysts pointing at a possible drop into the ¥140s during the year. A $100/month plan is often written as “¥15,000/month,” but depending on FX it can swing into the ¥14,000–16,000 range.
Put these three together and the correct answer to “Claude Code, how much per month?” becomes “it depends on how you use it.” In the marketing world, that’s also usually the right answer to “so what’s the actual price?” It’s not a brush-off — it’s a structure where if you fit your usage into one of a few templates, you can answer it instantly.
What I do is sort the person’s usage into one of three buckets — “personal trial,” “corporate PoC,” “production rollout” — and estimate scenario by scenario.
The 5 plans that include Claude Code, in 60 seconds
There are currently six billing entrances for Claude Code (counting Free). The table below is the overview.
| Plan | Monthly (USD) | Approx. JPY ($1=¥150) | Claude Code access | Target |
|---|---|---|---|---|
| Free | $0 | ¥0 | No | Chat trial only |
| Pro | $20/month | ~¥3,000 | Yes (light) | Personal trial / weekend use |
| Max 5x | $100/month | ~¥15,000 | Yes (medium) | Personal professional use |
| Max 20x | $200/month | ~¥30,000 | Yes (heavy) | Always-on personal agents |
| Team Premium | $125/seat/month ($100 annual) | ~¥18,750/seat | Yes | Small corporate teams |
| Enterprise | Custom | Custom quote | Yes | Org-wide rollout, governance requirements |
Sources: Anthropic official pricing / Claude Code usage guide (as of May 2026)
Three things to note here.
First, Free does not include Claude Code. Chat only. “Let me try it free first” doesn’t actually exist — and that’s the first stumbling block for everyone searching.
Second, Pro $20 does include Claude Code, but with the 5-hour rolling window and a weekly cap. If you run heavy workloads multiple times a day, you’ll hit the cap fast. For “I’ll touch it on weekends” usage, $20 is plenty.
Third, the “5x” and “20x” in Max refer to roughly 5x and 20x the Pro usage allowance. Within a 5-hour window, Max 5x can process around 88,000 tokens and Max 20x around 220,000 tokens (Verdent research; varies with operating conditions). The dividing line between these tiers is “are you going to run agents all day every day?”
For corporate plans, Team’s Premium seat ($125/seat/month) includes Claude Code access. The Standard seat ($25/seat/month) does not include Claude Code, so the assumption is that developers and technical staff get Premium seats (official corporate plan announcement).
The API (via Anthropic Console) is a separate track. No monthly subscription cap — you pay per token. Sonnet 4.6 is $3 input / $15 output, Opus 4.7 is $5 input / $25 output, Haiku 4.5 is $1 input / $5 output (all per million tokens).
From here, we’ll plug these into the three scenarios and work out the monthly numbers.
Scenario 1: Trying it personally — when “Pro $20 is enough” vs. when “you should jump to Max $100”
The most common question by far: “I want to try it personally, where do I start?”
Here’s the bottom line. If you’re using it weekends only / under 10 hours a month, Pro $20 is enough. If you’re touching it 5 days a week / 40+ hours a month, Max 5x $100 will be both cheaper and faster in the end.
Three conditions where Pro $20 is sufficient
First: your daily Claude Code use stays within 1–2 sessions. Second: each session is mostly under 20 minutes (small code edits, short refactors, adding a function). Third: “wait a few hours when I run out” is acceptable to you.
If all three apply, there’s not much point in paying ¥15,000 for Max. While the 5-hour Pro window resets, you go make some coffee — total cost stays around ¥3,000/month. Hobbyist programmers, side-project engineers, and people who dip in between work tasks fit this zone.
Signals that you should “move up to Max 5x”
The signals to move up are also three:
First: during work hours, you’re invoking Claude Code 3+ times a day. Second: you’re increasingly running tasks that span 10+ files in one session. Third: “running out of quota and waiting hours” is starting to interrupt your flow.
That’s the moment to upgrade to Max 5x. Max 5x’s 5-hour window is around 88,000 tokens — roughly 5x Pro. You also get Opus access, so it scales to complex requirement design and architectural support.
There’s a developer case where someone ran 1B tokens × 8 months (8B tokens total) on Max $100 — work that would have cost ~$15,000 at API rates ran for $800 (NxCode analysis). It’s an extreme case, but it shows how dramatically the flat-rate Max wins the moment you cross into professional use.
Max 20x $200 means “always-on”
You move to Max 20x $200 when you want to keep Claude Code running on your machine as a persistent agent. Use cases: parallel work across multiple terminals, overnight batch fixes, 24/7 logging monitoring. If you keep banging into Max 5x’s ceiling, or your high-priority processes are getting throttled, the upgrade pays for itself.
If you’re considering it “just in case,” skip it. Going from Max 5x with quota to spare to 20x just doubles your cost without changing anything you feel.
In personal scenarios, 80% of the people who can’t move are stuck because they jumped straight to evaluating Max 20x and decided “¥30,000/month is too expensive.” Start with Pro $20, upgrade to Max 5x when you need to. Two stages is enough.
Scenario 2: Running a corporate PoC — Team Premium or API Console? If torn, follow this order
Next, the corporate question: “We want 3–10 people running a PoC for 1–2 months.”
Two options: Team Premium ($125/seat/month) or API Console (metered).
When to choose Team Premium
If you’re handing Claude Code out evenly to 5–20 people, Team Premium is the only call.
Three reasons. First: per-seat caps are clear, so monthly budget is predictable. Second: the admin console gives you usage logs and cost analytics (official corporate plan announcement). Third: SSO, audit logs, and other governance features for Claude Code are included.
You can start at 5 × $125 = $625/month (~¥93,750). With annual billing, the per-seat rate drops to $100, bringing the monthly equivalent down to roughly ¥75,000.
When to choose API Console
API Console fits when a small group (1–3 specific people) is running heavy workloads in a focused PoC.
Say 1 person uses 1M tokens/month on Sonnet 4.6, with a 50/50 input/output split — that’s about $9/month (¥1,350). Three people: $27/month (¥4,000). Way cheaper than Max’s $100/month, but with no cap, budget control gets harder.
The trick with API Console PoCs is to use Anthropic’s prompt caching (up to 90% off) and Batch API (50% off) from day one (Anthropic official pricing). If you’re sending the same long system prompt every call, caching alone can cut actual cost by 30–50%.
Corporate PoC — if torn, run it in this order
The order I always recommend:
Stage 1 (Weeks 1–2): On API Console, have 2–3 key people verify “what can Claude Code do for my actual work?” Budget around $30–$100/month (author estimate).
Stage 2 (Weeks 3–6): If the first stage shows “yes, worth giving to everyone,” contract Team Premium for 5–10 people and expand the PoC. $625–$1,250/month.
Stage 3 (Week 7+): With ROI data in hand, decide: continue Team, evaluate Enterprise, or pull out.
Following this order avoids the classic failure mode of “we contracted a big Team Premium block and only 3 people actually used it.” The first wall in corporate adoption is “buying seats before deciding who uses them.”
The 3 vendors covered in my May 3 article on Claude Code corporate adoption becoming a market — NEC, Givery, DigiRise — bring their value in the post-PoC full-rollout phase. You can handle the plan-selection step entirely in-house.
Scenario 3: Production rollout / org-wide deployment — Enterprise vs. “API at scale”
The third scenario: you’ve finished the PoC and are moving to production.
Two options: Enterprise (flat per seat) or “API at scale” via Anthropic Console (metered + volume discount).
When Enterprise fits
Enterprise fits when you’re handing Claude Code to hundreds or tens of thousands of staff. NEC’s deployment to all 30,000 group employees (Anthropic announcement), or AR Advanced Technology rolling it out to all engineers and consultants (see my April 19 article), are at this layer.
Enterprise pricing isn’t published. It’s a custom quote covering seat count, usage, SSO/SCIM and other requirements. As a guideline, it’s said to be a Premium-equivalent seat rate plus a premium for governance features. The Enterprise contact form on Anthropic’s site is the starting line.
Three triggers for considering Enterprise. First: more than 50 users. Second: audit logs, SSO, and data governance are requirements. Third: vendor-side support SLAs are needed. When all three line up, evaluate Enterprise.
When API at scale fits
API at scale fits when you’re embedding Claude Code into in-house tools or agents. Plugging a review agent into your CI/CD, embedding coding assistance into an internal bot — those are the canonical use cases.
For API at scale, Sonnet 4.6 base rates are $3 input / $15 output. At 1B+ tokens/month there’s room to negotiate volume discounts. Standard practice is to budget assuming prompt caching + Batch API will cut actual cost by 30–50%.
For a 100-engineer org running 1M tokens/day × 100 people = 300M tokens/month, with a 50/50 split, that’s $3,000/month (¥450,000) — plus cache optimization brings actual spend into the $1,500–2,000 range. That’s the production-operations ballpark I’ve seen (varies with operating conditions; author estimate).
Three basic moves to optimize cost in production
I’ve seen multiple companies enter production and find their costs at 2x the projection 1–3 months in. The shared causes and fixes:
First: defaulting to “Opus for everything.” Opus 4.7 is 1.7x the input rate and 1.7x the output rate of Sonnet 4.6. Day-to-day tasks on Sonnet, Opus only when truly needed, lightweight classification on Haiku — model routing is the basic move.
Second: not using prompt caching. Sending long system prompts and codebase context every call inflates input tokens forever. Designing operations around the 90%-off cache discount is the standard (Anthropic official: Manage costs).
Third: not using Batch API for “doesn’t need to be real-time” work. Overnight batches, bulk classification, report generation — Batch API (50% off) handles them fine.
Bake these three into operational design from day one and your first production month lands inside budget.
A 30-second decision flow so you don’t get lost
We’ve covered 5 plans × 3 scenarios. Here’s a flow that lets you decide your starting plan in 30 seconds.
Q1: Who’s going to use it?
- Just me alone → go to Q2
- 3–10 people at the company → go to Q3
- Whole company (hundreds+) → go to Q4
Q2: How often per week will you call Claude Code?
- Weekends only / under 10 hr/month trial → start with Pro $20/month (~¥3,000)
- Daily on weekdays / 40+ hr/month professional use → go straight to Max 5x $100/month (~¥15,000)
- Always-on, multiple parallel agents → Max 20x $200/month (~¥30,000)
Q3: Are you handing it to everyone evenly, or focused trial with a few people?
- Evenly to everyone → contract Team Premium $125/seat/month for the seats you need
- Focused trial with 1–3 people → API Console (Sonnet 4.6, ~$30–$100/month)
Q4: For org-wide rollout, do you have governance requirements?
- SSO, audit, SLA requirements → contact Enterprise
- Embedding into in-house tools/agents → API at scale (negotiate volume)
The point of this flow is “don’t pick the expensive plan first.” Try Pro and move up to Max; PoC on API Console and expand to Team; advance to Enterprise as needed. Stage your moves and a few thousand yen of trial-and-error gets you to your right answer.
In practice, almost every failure I’ve seen around me is one of two patterns: “contracted Max 20x and canceled within two weeks” or “bought 20 Team Premium seats and half went unused” (author observation). It’s because pricing is hard to read that starting at the smallest entrance and stepping up gradually drives total cost down.
Wrap-up: three things to do this week
Action 1 (30 minutes today): Open the official Anthropic pricing page and use the decision flow above to pick exactly one entry plan. Anything but Free.
Action 2 (this week): If personal, sign up and actually use it for a week. If corporate, pick 2–3 PoC kickoff members and issue API Console accounts (5 minutes with an admin Google or Microsoft account).
Action 3 (put a date on your calendar a month out): A month from now, look at usage logs and decide “upgrade, continue, or switch plans.” Setting the review date in advance is the single most effective move for cost optimization.
“I can’t move because I can’t read the pricing” is the biggest wall to trying Claude Code. If this article helps you over that wall, that’s enough.
Pricing changes. Always confirm current numbers on the official Anthropic pricing page and the Claude Code cost management docs.
If you’re considering corporate adoption, my May 6 article covers “onboarding design that doesn’t stall in the first 30 days” — once you’ve picked your entry plan. Use it as a blueprint for the phase right after plan selection.
Sources & references
AIを使いこなせない方は、この先どんどん差がつきます。僕はAIエージェントを毎日動かして、壊して、直して、また動かしてます。そういう泥臭い実践の記録をここに書いてます。理論は他の方にお任せしました。僕は動くものを作ります。朝5時に起きてウォーキングしてからコードを書くのがルーティンです。
