Claude Code · Claude Max alternative

A Claude Max alternative for Claude Code: pay per token

Claude Max is a good deal if you use it hard every week. If you keep hitting its 5-hour or weekly cap, or pay for months you barely code, a per-token Claude API can fit better: Claude Code keeps working the same way, and you pay only for the tokens you use. This page compares the two honestly, including where Max still wins.

Last checked 2026-10-10

Get an API key See Claude model prices

Cost: Max vs a per-token API

A plan costs the same every month; per-token billing depends on how many days you code and how heavily. The break-even table shows how many active days at each usage level equal one month of each plan, using live token prices.

How many active days of per-token usage equal one month of a Claude plan
Plan (USD / month)	Light days (Claude Sonnet 5.5) at list price / at aicodewith Economy channel	Moderate days (Claude Opus 5.5) at list price / at aicodewith Economy channel	Heavy days (Claude Opus 5.5) at list price / at aicodewith Economy channel
Claude Pro ($20)	4.2 days / 21.9 days	2.1 days / 11.0 days	0.9 days / 4.5 days
Claude Max 5x ($100)	20.8 days / 31+ days	10.4 days / 31+ days	4.3 days / 22.7 days
Claude Max 20x ($200)	31+ days / 31+ days	20.8 days / 31+ days	8.6 days / 31+ days

Read it as: if you have fewer active days per month than the number shown, per-token billing costs less than that plan. If you work more days than that and stay inside the plan's allowance, the plan is usually cheaper. "31+ days" means per-token billing costs less than the plan even if you code every day. Heavy days may not fit inside a plan's 5-hour and weekly allowance at all. Plan prices from claude.com/pricing; token prices are live.

Example Claude Code costs per day and per month
Usage	Model	Tokens per active day	At Anthropic list price / day	22 days / month	aicodewith Economy channel / month
Light	Claude Sonnet 5.5	15M cache read, 0.6M cache write, 0.15M input, 0.15M output	$4.80	$106	from $20.06
Moderate	Claude Opus 5.5	15M cache read, 0.6M cache write, 0.15M input, 0.15M output	$9.60	$211	from $40.13
Heavy	Claude Opus 5.5	40M cache read, 1.2M cache write, 0.3M input, 0.4M output	$23.20	$510	from $96.98

Token mixes are illustrative (most Claude Code tokens are cache reads). Run /cost in Claude Code to see your own numbers. More detail on the usage limits page.

Claude model prices in USD per 1M tokens
Model (USD per 1M tokens)	Anthropic list price, input / output	Cache read / 5-min cache write (list)	aicodewith Economy channel, input / output
Claude Opus 5.5	$4.00 / $20.00	$0.20 / $5.00	from $0.76 / from $3.80
Claude Sonnet 5.5	$2.00 / $10.00	$0.10 / $2.50	from $0.38 / from $1.90
Claude Haiku 5.5	$0.10 / $0.50	$0.01 / $0.125	from $0.019 / from $0.095

"From" prices are based on the Economy channel, the lowest-priced Claude channel currently listed (about 19% of Anthropic list price). Other channels cost more; the rate depends on the channel you choose, and the models page is always the source of truth.

Full cost math on the usage limits page

Does a per-token API have limits too?

Yes. It does not use a plan's 5-hour session allowance or weekly allowance, but it is not limitless. Rate and concurrency limits apply to every account, and requests above them return 429. We don't publish fixed per-minute numbers here because they can change; if you plan heavy parallel workloads, check the docs or contact support first.

	Claude Max	Per-token API on aicodewith
Session cap	Allowance resets every 5 hours	None; you pay per token
Weekly cap	Yes, amount not published	None; your balance is the limit
Rate and concurrency limits	Applies	Applies; 429 when exceeded
Your own budget controls	Usage credits with an optional monthly cap	Daily, monthly or total spend cap per API key
When you run out	Wait for the reset or buy usage credits	Top up; nothing is charged automatically

How Max limits work Docs

What you change in Claude Code

Two settings: the endpoint and your key. Add them to ~/.claude/settings.json (Windows: %USERPROFILE%\.claude\settings.json), restart Claude Code, and run /status to confirm the endpoint is api.aicodewith.ai.

~/.claude/settings.json

{
  "env": {
    "ANTHROPIC_AUTH_TOKEN": "YOUR_API_KEY",
    "ANTHROPIC_API_KEY": "",
    "ANTHROPIC_BASE_URL": "https://api.aicodewith.ai"
  }
}

ANTHROPIC_BASE_URL is the bare domain, with no /v1. To go back to your Max plan, remove the env block and run /login.

Step-by-step Claude Code setup API key setup in detail

What Max includes that an API key does not

An API key only covers API requests, such as the ones Claude Code makes. A Max plan is a whole Claude subscription. If you switch completely, you lose:

Chat on claude.ai and in the Claude desktop and mobile apps.
Projects, and other features that live inside the Claude apps.
A predictable bill: per-token cost goes up on heavy days.

What carries over: Claude Code itself, its IDE extensions, /model switching between Claude models, subagents, MCP servers, CLAUDE.md and your settings.

Who it suits, and who should stay on Max

A per-token API suits you if	Stay on Max if
You hit the 5-hour or weekly cap and lose working time	You use most of your allowance every week and rarely hit the cap
Your usage is uneven: sprints, then quiet weeks	You code most days at a steady pace
You need separate budgets per teammate, project or CI job	One person, one machine
You mostly use Sonnet or Haiku	You run Opus all day
You only use Claude through Claude Code	You also rely on claude.ai chat and Projects

A common middle ground: keep Max, and point Claude Code at an API key only on the days you hit the cap.

On Claude Pro? Your options

Pro has a smaller allowance than Max, so Claude Code users hit it sooner. Before upgrading to Max, compare these:

Stay on Pro and save allowance: /clear between tasks, /compact mid-task, and Sonnet instead of Opus for routine work.
Turn on usage credits in your Claude account for occasional overflow; they are billed at Anthropic's standard API rates.
Keep Pro for chat and switch Claude Code to a per-token API key, so coding no longer draws on the Pro allowance.
Upgrade to Max if you would use most of its allowance every week.

Use Claude Code without a subscription

FAQ

What is the best Claude Max alternative for Claude Code?

If you want to keep using Claude Code with Claude models, the closest alternative is a per-token Claude API: set ANTHROPIC_BASE_URL and an API key, and pay only for the tokens you use. If you want a different tool, see the Claude Code integration guides for Codex, OpenCode, Cline and others.

Is a per-token API cheaper than Claude Max?

It depends on how much you use. If you have fewer active days per month than the break-even table shows, it is cheaper. If you use most of the Max allowance every week, Max is usually cheaper.

Does the per-token API have rate limits?

Yes. There is no 5-hour or weekly allowance, but rate and concurrency limits apply and requests above them get a 429 response. You can also set spend caps per API key.

Do I get the same Claude models?

You choose the model in Claude Code with /model. Claude Opus 5.5, Sonnet 5.5 and Haiku 5.5 are available; the models page lists every model with live prices.

Can I keep my Max plan and use an API key only sometimes?

Yes. Add the API key settings when you hit the cap, and remove them and run /login to go back to your plan.

What do I lose by leaving Max?

Chat on claude.ai and in the Claude apps, Projects, and a fixed monthly bill. Claude Code itself works the same way.

How do I pay, and can I get a refund?

International accounts pay in USD by card through Stripe; mainland China accounts pay in CNY through Alipay. Unused balance can be refunded within 90 days of the top-up, with a 10% fee deducted.

Is aicodewith affiliated with Anthropic?

No. aicodewith is an independent API platform. Claude, Claude Max and Claude Code are trademarks of Anthropic.

Ready to try it?

Billing is per token from a prepaid balance. Rate and concurrency limits apply, and you can set daily, monthly or total spend caps on each API key.