Pro/Max plan or pay-as-you-go API? Picking the right way to pay for Claude

Same Claude, same models — but two completely different ways to pay: a subscription (Pro, Max, Team) with a flat monthly fee, or the API, billed per token. Pick wrong and you either overpay or hit a wall right when you need it — and using the wrong one for the wrong purpose can breach the terms. This article compares them in detail so you (or your team) choose correctly.

Two billing models

Current subscription plans

A few common misunderstandings:

• Claude.ai, Claude Desktop and Claude Code share one limit, which resets every 5 hours. Heavy chatting on the web means Claude Code runs out sooner.
• Max plans also have a weekly cap across all models, resetting at a fixed time each week.
• Anthropic doesn’t publish token or message counts per plan, so an exact break-even against the API can’t be calculated — only estimated.
• Per “unit of Pro usage”, Max 5x costs the same as Pro ($20), while Max 20x costs half ($10).

Reference API prices (USD per million tokens)

Prices and limits change over time — always check the official pricing pages before presenting a budget.

Two quick calculations

Light use: the API is far cheaper

An internal tool calls Sonnet 5.5 about 40 times per working day, each with ~2,000 input and ~500 output tokens. Over 21 working days that’s 840 requests:

Input : 840 × 2,000 = 1.68M tokens × $2  = $3.36
Output: 840 × 500   = 0.42M tokens × $10 = $4.20
Total ≈ $7.56 / month   (cheaper than the $20 Pro plan)

Heavy use: the subscription wins

One author measured a single intense Claude Code day (2,979 turns) at roughly $220 at Opus 5.5 API list prices — more than a whole month of Max 20x ($200). For developers running agents all day, a subscription is almost always cheaper.

Terms: never use a subscription as a backend

Since 19 February 2026, Anthropic’s documentation makes it explicit: OAuth tokens from Free, Pro and Max plans are only for Claude Code and Claude.ai. Using them in other products, tools or services — including products built on the Agent SDK — breaches the terms. Personal experimentation with the Agent SDK or claude -p is still encouraged; but when you build a business on it, use an API key from the Claude Console or a supported cloud provider.

Deciding in 4 steps

1. Who is calling? A person typing → subscription. A machine calling automatically, or serving your users → API.
2. Measure for two weeks. In Claude Code, check /status to see how often you hit the limit. For the API, check the Usage page in the Console.
3. Apply the rule of thumb:
   • Estimated API cost well under $20/month → use the API.
   • Estimated API cost well over $200/month → pick a subscription.
   • In between → try Pro for a month and reassess.
4. Upgrade by how often you hit the cap: rarely or a few times a month → stay on Pro. Several times a week, and waiting disrupts your work → Max 5x. Still capped on Max 5x → Max 20x.

A combined setup for teams

Most real software teams use both:

• Subscriptions per developer/PM: Claude Code and Claude.ai for daily coding, review and documentation.
• API for the product and automation: AI features in the app, CI pipelines, overnight batch jobs, customer-support bots.

How to apply it

• Developers: If you’re using subscription tokens in scripts or home-grown tools that serve other people, switch to an API key now — before the account gets locked mid-project.
• Tech Leads: Standardise: one API key per environment (dev, staging, prod), each with spend limits and alerts in the Console.
• PMs: Split the budget into two lines — “team tooling” (subscriptions, fixed per head) and “product operations” (API, variable with traffic). They have different owners and different forecasting methods.