# Claude Code automation billing: subscription limits and API credits
> Distinguish Claude subscription limits, included Max and Team API credits, and optional paid API usage; estimate costs from measured runs.
Source: https://ailoot.ai/blog/en/claude-code-pricing-automation-billing/
Canonical: https://ailoot.ai/blog/en/claude-code-pricing-automation-billing/
Language: en
Publisher: Ailoot (https://ailoot.ai/)
Translation: guide-044
Revision: sha256-094bbdc35ef74ef15aeb9d60232e791578445f6b31ea811117574f407ddd8a18
Key takeaway
The June 15, 2026 change described in the original announcement was paused. Anthropic’s support update says Agent SDK, `claude -p`, and third-party apps can still use subscription limits. The October 7 update introduced separate monthly API credits for eligible Max and Team plans; those are an included allowance, not a second mandatory subscription bill. Check which identity authenticates each run before estimating its cost.
Model connections and API usage are described in the [Ailoot documentation](https://ailoot.ai/docs/).
## How are Claude Pro and Max automation charges changing?
The original May announcement proposed moving Agent SDK and programmatic usage to a separate monthly pool from June 15. That proposal did not take effect: [Anthropic’s support update](https://support.claude.com/en/articles/15036540-use-the-claude-agent-sdk-with-your-claude-plan) says it was paused.
The practical distinction is now authentication. A run signed in with your Claude plan can use plan limits. A run authenticated with a Claude Console API key uses that organization’s API balance. Interactive chat and terminal work remain on plan limits. Do not classify a bill merely by whether the process runs unattended.
For the separate October API allowance and its current eligibility, use the [monthly API credits guide](https://support.claude.com/en/articles/17154008-monthly-api-credits-for-max-and-team-plans). The abandoned June allowance is not available to claim.
## A simple test: which balance does your run use?
List your GitHub Actions, scheduled scripts, Agent SDK applications, and interactive sessions. For each, record its authentication method and launch surface. A subscription login and an API key are different billing paths even when they invoke the same model.
Use one small development task and inspect its usage record in the relevant account. Do not infer an API-credit deduction solely from seeing `claude -p` in a command: the official credits guide distinguishes a self-run API-key process from Claude Code usage started by GitHub Actions, IDE extensions, or the desktop app.
Typical cases worth checking are overnight PR reviews, recurring research scripts, n8n wrappers, and third-party tools. Keep their credentials and spending limits explicit before enabling unattended work.
## How much credit you get and what's included
Under the October API-credit policy, Max 5x includes $100 monthly, Max 20x $200, and Team includes $20 per Standard seat or $100 per Premium seat, pooled up to $500. Free, Pro, and Enterprise are ineligible. These replace the abandoned June proposal as the current reference.
API credits are granted to one linked Console organization. They are separate from Claude plan limits and do not cover interactive Claude Code or its extra usage. Use the official eligibility and launch-surface rules before budgeting a particular workflow.
For a cost estimate, calculate input tokens and output tokens separately at the model’s published rates. The sample calculations below use the article’s dated Opus 4.8 prices of $5 per million input and $25 per million output tokens; they are arithmetic examples, not a measurement of your runs.
A sample 30–50K-input/15–20K-output run costs about $0.50–$1. At that illustrative range, $100 covers roughly 100–200 runs and $200 roughly 200–400. Long runs consume more; short ones consume less.
## What happens when the credits run out
Monthly API credits expire each billing cycle. If no other credits are available, API requests stop. Purchased credits or auto-reload can keep API usage running; invoiced organizations follow their billing agreement. This does not create an extra charge to the Claude plan itself.
Check the Console organization’s balance and reload settings before unattended work. A notification is not a hard spending cap. Keep a budget and monitor actual usage instead of assuming a monthly allowance guarantees uptime.
## 3 strategies: what to choose for your usage
### Strategy 1: Keep the existing plan path
If occasional development tasks fit your subscription limits, the paused June proposal is not a reason to buy a second plan or migrate credentials. Verify the actual usage path first.
### Strategy 2: Use an eligible API allowance with a bounded backup balance
For eligible Max or Team API work, check the linked Console organization, remaining allowance, and optional purchased-credit or auto-reload settings. Estimate the cost of a representative run and set a budget before enabling extra spending.
### Strategy 3: Use a standalone API budget
For production automation or a shared team bill, compare a Console API setup with your existing plan path. Keep or cancel your personal subscription based on interactive needs; an API key is a separate usage budget. Test the integration in development before changing a working production credential.
There is no universal job-count threshold at which a migration becomes cheaper. Use measured run costs and your completion rate.
## Calculator: how many runs will fit?
Key takeaway
Calculate it this way: how many tokens one of your runs uses on average → multiply by the model price → divide your credit by that amount. That gives you the maximum number of runs per month.
Here is a sample budget calculator. The $20 column is a hypothetical budget, not a Pro entitlement; Pro is not eligible for the October API credits. The figures use Opus 4.8 (\$5 per million input tokens, \$25 per million output tokens—current at the time of publication; check [Anthropic's pricing page](https://platform.claude.com/docs/en/about-claude/pricing) if you're reading this much later).
| Task type | Tokens per run | Cost per run | Hypothetical $20 budget | Max 5x ($100) | Max 20x ($200) |
|----|----|----|----|----|----|
| Short code review (PR with 3–5 files) | 10K in + 5K out | \$0.18 | ~110 runs | ~550 | ~1100 |
| Average agent (research or module refactoring) | 30K in + 20K out | \$0.65 | ~30 runs | ~150 | ~300 |
| Long overnight run (Claude analyzes architecture for hours) | 100K in + 50K out | \$1.75 | ~11 runs | ~57 | ~114 |
| Heavy workflow (Claude assigns itself subtasks) | 200K in + 100K out | \$3.50 | ~5 runs | ~28 | ~57 |
If you're on **Sonnet**, multiply the figures by about 1.5—the model is 1.5–2 times cheaper. On **Haiku**, it's 5–10 times cheaper, so you can fit proportionally more.
A simple estimate for your own project:
1. Check the [Agent SDK cost-tracking page](https://platform.claude.com/docs/en/agent-sdk/cost-tracking) to see how much you're spending per week now.
2. Multiply by 4.3 to get the monthly spend.
3. Compare that with the verified balance and eligibility for your chosen billing path.
If the allowance is too small, compare a bounded purchased-credit budget with a standalone API setup. Use actual costs rather than a fixed ratio as the decision rule.
## What to check now
1. Check your plan, authentication method, launch surface, and remaining usage in the corresponding account.
2. If eligible for API credits, follow the current official linking instructions and check the credited organization. Do not wait for a June email or assume the abandoned allowance can still be claimed.
3. Estimate monthly usage from measured runs, including failed attempts and rereads.
4. Decide whether usage should stop at the available balance or whether to add a bounded paid backup. Verify the result in a development run.
5. For a team, ask the administrator who controls API keys and spending. Do not assume an individual seat’s allowance is private or transferable.
The old June 14/15 deadlines are historical and are not setup steps for a reader now.
## Alternatives if the credits aren't enough
Key takeaway
If your automation volume won't fit in the credit and you're nervous about enabling overflow billing, there are competitors. Compare other tools by permitted automation usage, real task cost, and current plan terms rather than treating advertised allowances as interchangeable.
If your automation is heavy, it may be worth looking at competitors.
**OpenAI Codex CLI**—included with a ChatGPT Plus subscription for \$20 a month. No separate credit pool and no changeover date. Supports `/hooks`, `/goal`, permission profiles. For typical tasks (code review, refactoring based on an example, tests), it delivers comparable quality.
**Cursor Ultra**—\$200 a month gives you a \$400 “programmatic envelope” at API rates. The advertised allowance is not enough by itself to establish a better deal: compare eligible usage, model prices, completed tasks, and current terms. Anthropic’s API credits and subscription limits are separate.
**Cursor Composer 2.5**—an inexpensive model based on Kimi K2.5, at \$0.50 per million input tokens and \$2.50 per million output tokens. Suitable as an “inexpensive workhorse” for high-volume, long tasks that don't need top-tier intelligence.
I reviewed the alternatives in detail:
- [Codex vs Claude Code comparison](https://ailoot.ai/blog/en/codex-vs-claude-code-choice/)
- [Cursor vs Claude Code](https://ailoot.ai/blog/en/cursor-vs-claude-code/)
- [How to run both overnight with /goal](https://ailoot.ai/blog/en/goal-mode-claude-code-codex/)
## Payment method for usage credits
Check [Console billing](https://platform.claude.com/settings/billing) before buying extra API credits. Claiming the included API allowance does not itself require adding a payment method; additional paid usage has its own billing requirements.
## Interactive use and automation: keep the paths distinct
Chat, interactive Claude Code, and Cowork use their plan limits. Agent SDK or `claude -p` can also use subscription limits when signed in with the plan. API-key work uses Console billing, subject to the allowance’s eligibility rules.
Skills, subagents, MCP, plugins, and hooks do not define a separate bill by themselves. Record how the containing process is authenticated and launched. Preserve a working setup until a small development run demonstrates the intended usage and billing path.
## A separate historical notice: June model retirements
Key takeaway
The original June notice also discussed removing two outdated model IDs from the API—`claude-sonnet-4-20250514` and `claude-opus-4-20250514`. Check any hard-coded model IDs against the current deprecation page before changing an application. Also, the TypeScript package was renamed: `@anthropic-ai/claude-code` is now `@anthropic-ai/claude-agent-sdk`.
This is a separate historical migration notice, not evidence that the paused billing proposal took effect. Verify model availability and package migration requirements independently before editing a working application.
**What is being removed from the API:**
- `claude-sonnet-4-20250514`
- `claude-opus-4-20250514`
These IDs and their suggested replacements belonged to the original notice. Do not copy a historical replacement as a current recommendation: select an available model from Anthropic’s [current model documentation](https://platform.claude.com/docs/en/about-claude/models/overview) and test it in development.
**What was renamed:**
- TypeScript: `@anthropic-ai/claude-code` → `@anthropic-ai/claude-agent-sdk`
- Python: `claude-code-sdk` → `claude-agent-sdk`
The original notice said the old package names redirected to the renamed packages. Check the current package migration guidance before relying on that historical behavior.
**What to do now:**
1. ### grep your scripts
`grep -r 'claude-sonnet-4-20250514\|claude-opus-4-20250514\|claude-code-sdk\|@anthropic-ai/claude-code' .`—if anything turns up, update it.
2. ### Update package.json or requirements.txt
Compare the installed package and hard-coded model IDs with the current migration and availability documentation; change only what that comparison requires.
3. ### Run your tests
Run your Agent SDK tests with any changed names before applying them in production. The old June deadline is not a current setup step.
## Five common questions
**Was the June allowance a second required monthly payment?**
No. It was announced as an included credit allowance, and the proposed change was later paused. Do not add its amounts to the subscription price.
**Can a Pro user claim the October API credits?**
No. The current guide lists eligible Max and Team plans. Pro and Enterprise are not eligible.
**Do credits roll over?**
No. Check their expiry in Console and budget within the billing cycle.
**Should I wait for the old June email?**
No. Use the current billing page and official eligibility instructions; the abandoned Agent SDK credit is unavailable.
**Does an existing API-key integration have to change?**
The announcement alone does not require a new key. Check the organization balance and billing path, preserve working credentials, and test any justified change in development.
## What to take away
There are three separate things to check: subscription limits, an eligible included API allowance, and optional paid API usage. The June change was paused; it is not a second mandatory bill.
Record authentication and launch surface, measure a representative run, and compare the resulting cost with the verified balance. Clean context and organized project instructions can reduce rereads, but measure that saving rather than promising a fixed reduction.
### Sources
- [Use the Claude Agent SDK with your Claude plan - Anthropic Support](https://support.claude.com/en/articles/15036540-use-the-claude-agent-sdk-with-your-claude-plan) — June proposal paused; October update checked October 19, 2026
- [Monthly API credits for Max and Team plans](https://support.claude.com/en/articles/17154008-monthly-api-credits-for-max-and-team-plans) — current eligibility, balances, linking, launch-surface rules, and optional extra usage, checked October 19, 2026
- [Agent SDK overview - Claude Code Docs](https://code.claude.com/docs/en/agent-sdk/overview)
- [Track cost and usage - Claude API Docs](https://code.claude.com/docs/en/agent-sdk/cost-tracking)
- [Claude Pricing](https://claude.com/pricing)
- [Billing - Claude Developer Platform](https://platform.claude.com/settings/billing)
You can continue working with AI in [Ailoot](https://ailoot.ai/).
## Questions about Claude Code automation billing
### Do the announced credit amounts add a second mandatory bill?
No. The June proposal described an included allowance and was paused. Current eligible Max and Team API credits are separate from plan limits; extra API spending depends on the linked Console organization’s purchased credits, reload settings, or invoicing.
### How do I determine which balance a scheduled run uses?
Check its authentication and launch surface, then inspect a small development run’s usage record. A subscription-signed-in process and a Console API-key process can use different billing paths; the presence of `claude -p` alone does not prove an API-credit deduction.
## Ailoot · The company and its project portfolio
Ailoot combines established operating experience, a large client base, stable service and ongoing KPI control.
The company and its project portfolio: millions of customers; monthly sales in the billions.
Ailoot brings that operating experience to API access for AI features in customer products, agents, applications and automations.