Choosing
Claude Code on a subscription, the API or a relay?
Claude Code is free to install; what costs money is the model behind it. There are three ways to pay: a Claude subscription, the Anthropic API billed by the token, and a third-party relay. They differ in price, but also in usage limits, billing and where they work. The official details below come from Anthropic's pricing page and the Claude Code docs, checked on October 30, 2026. CoreCode is itself a relay, so on relays we stick to what you can check yourself.
- Subscriptions: Pro is $20 a month ($17 billed annually); Max starts at $100 a month for 5 or 20 times Pro's usage. Usage is counted in rolling 5-hour windows plus weekly limits, and Claude Code shares one pool with chat.
- API: no monthly fee, billed by the token; Sonnet 5.5 is $2 per million input tokens and $10 per million output tokens. Anthropic reports an enterprise average of about $13 per developer per active day.
- Anthropic's supported regions exclude mainland China, Hong Kong and Macau, so developers there face account, network and payment hurdles with both subscriptions and the API.
- Relays bill by the token and take local payment, but whether the upstream is really Anthropic and whether caching keeps hitting are things to verify yourself.
The three side by side
| Claude subscription | Anthropic API | Third-party relay | |
|---|---|---|---|
| How you pay | Fixed monthly: Pro $20, Max from $100 | Per token used | Per token used, usually discounted |
| Usage limits | A rolling 5-hour window, plus weekly limits | Rate limits by usage tier; no total cap | Set by each relay |
| Billing detail | Usage bars and a breakdown, not a per-token bill | Metered per request in the Console | Depends on the relay; good ones list every request |
| Main risk | Waiting for the window to reset when the allowance runs out | Costs climb fast when a long session's cache misses | A substituted upstream, opaque billing, a relay that disappears |
Claude subscriptions: Pro and Max
- Pro: $20 billed monthly, or $17 a month billed annually ($200 up front). At least 5 times the usage of the free plan per 5-hour session.
- Max: from $100 a month, with 5 or 20 times Pro's usage, plus $100 to $200 a month in API credits.
- The free plan does not include Claude Code. Every paid plan does, and Claude Code draws from the same allowance as chat on the web and in the apps.
- Every plan has rolling 5-hour windows, and paid plans also have weekly limits. When the allowance runs out, paid plans can turn on usage credits and keep working at standard API rates.
A subscription suits someone who uses Claude Code daily at a steady pace: the cost is fixed and there is no bill to watch. The catch is that the allowance is not stated in tokens, so heavy use can hit a limit in the middle of the day.
The Anthropic API: pay per token
Sign Claude Code in with an API key and you pay for the tokens you use, with no monthly fee. The main models' prices are in Claude API pricing: Sonnet 5.5 is $2 input and $10 output per million tokens, Opus 5.5 $4 and $20, and a cache hit costs 5% of the input price.
What you spend depends on how you work. The Claude Code docs give enterprise figures: about $13 per developer per active day on average, $150 to $250 per developer per month, and under $30 per active day for 90% of users. Type /usage in Claude Code to see the current session's tokens and their cost at list price.
The API limits rate, not volume: Anthropic sets requests and tokens per minute by usage tier (Start, Build, Scale).
Third-party relays
A relay holds Anthropic accounts abroad. You point Claude Code's ANTHROPIC_BASE_URL at it and pay per token. It solves three problems for developers in mainland China: Anthropic's supported regions do not include it, overseas payment is hard to arrange, and the network.
Its risks come from that middle layer too: the upstream can be swapped for a cheaper model or a reverse-engineered endpoint, usage can be counted by the relay's own rules, and the cache can break when the relay keeps changing accounts. All of these can be checked:
- Run the checking script to confirm the response format, a cache read-back and the thinking signature, none of which can be faked.
- Go through the seven points in How to choose a relay: billing on Anthropic's terms, a per-request log, a public status record.
- Top up a small amount, run one real long session, and check that cache reads appear on the bill.
Choosing by how much you use it
- Occasionally, a few times a month: pay per token through the API or a relay, rather than a monthly fee you will not use.
- Every day at a steady pace, with an account and payment Anthropic accepts: a Pro or Max subscription is the least effort. Start with Pro and move to Max if you keep hitting limits.
- A team, or anyone who reconciles costs: per-token billing is clearer, with each request's usage and cost on record.
- In mainland China without overseas accounts: choose a relay you can verify, and try it for a week with the script and a small top-up.
Whichever you choose, a long session's cost depends mostly on the cache hitting; see Spending less.
FAQ
Can the free Claude plan use Claude Code?
No. Anthropic's pricing page lists Claude Code on paid plans only.
What happens when my subscription allowance runs out?
Wait for the 5-hour window to reset; turn on usage credits on a paid plan and continue at standard API rates; or switch to an API key and pay per token for a while.
Can Max's API credits pay for Claude Code?
The pricing page describes them as credits to build on the Claude Platform. They are separate from Claude Code's subscription allowance; check Anthropic's own terms for how they apply.
Sources
- Claude: Pricing (plan prices, usage multiples, 5-hour windows and weekly limits, Claude Code sharing the plan's allowance; checked October 30, 2026)
- Claude Code: Manage costs (enterprise averages, the
/usagecommand) - Anthropic: Pricing (API prices, rate-limit tiers)
- Anthropic: Supported regions