Claude Code Billing Change June 15, 2026

Update — June 16, 2026: this change was paused. Anthropic paused the billing split on the day it was due to take effect (June 15). The Claude Agent SDK, claude -p, and GitHub Actions usage still draw from your normal subscription limits — the separate per-user credit pool is not live. The Help Center article now opens with the pause notice. The model-ID retirements covered below did happen. For the current, post-pause picture, read Claude Code costs after June 15. The rest of this post documents what the change would have done, kept for reference.

Problem: A Help Center article landed in May telling you the way Claude Code charges automated work flips on June 15, 2026. That is tomorrow. Your claude -p cron job, the GitHub Action that reviews every PR, the third-party agent you wired in last month. All of those move to a separate monthly credit pool billed at full API rates.

Quick Win: Run this audit before anything else:

grep -r "claude-sonnet-4-20250514\|claude-opus-4-20250514\|@anthropic-ai/claude-code\|claude-code-sdk" .

Anything that comes back is broken on June 15. Two model IDs retire the same day the credit pool ships, and the SDK packages get renamed in lockstep. Three changes, one calendar square.

If you only use Claude Code in your terminal or IDE, close this tab. None of it touches you. Keep reading if you call claude -p, run Claude Code GitHub Actions, or build on the Claude Agent SDK.

What Changes on June 15, 2026

The Claude Code billing change on June 15, 2026 splits subscription usage into two pools. Interactive Claude Code in the terminal and IDE keeps using your existing subscription limits. Programmatic usage through the Claude Agent SDK, the claude -p command, the Claude Code GitHub Actions integration, and third-party apps draws from a new monthly Agent SDK credit billed at full API rates.

Source: Anthropic Help Center article 15036540.

What stays untouched:

• Claude Code in your terminal or IDE
• Claude.ai web, desktop, mobile chat
• Claude Cowork
• API key billing on the Claude Developer Platform

What now drains the new credit:

• Claude Agent SDK calls in your own Python or TypeScript projects
• The claude -p non-interactive command
• The Claude Code GitHub Actions integration
• Third-party apps that authenticate through your Claude subscription

Does This Affect You?

A short decision tree before you read further:

1. You only run claude in your terminal and never -p. You are fine. Stop here.
2. You wired Claude Code into a CI step, a cron, or a GitHub Action. Keep reading.
3. You build on the Agent SDK directly. Keep reading and budget for the package rename.
4. You manage an Enterprise Standard seat. Read the credit table below carefully. You get $0.

The New Agent SDK Credit by Plan

Sources: Anthropic Help Center, claudefa.st analysis, devtoolpicks.

How the Credit Works

Five mechanics worth committing to memory:

• Per-user. Cannot be pooled, transferred, or shared with teammates.
• Refresh monthly. No rollover.
• One-time opt-in required to claim the credit.
• Credits drain first. Extra usage kicks in after if you turned it on.
• Extra usage off and credit empty? Agent SDK requests halt until next billing cycle.

Sources: Anthropic Help Center, claudefa.st.

What It Actually Costs You

Run the math on a normal indie setup. One PR review per push. Four pushes per day. Thirty days. Sonnet 4.6.

Each PR review reads roughly 50k input tokens (the diff, surrounding context, your CLAUDE.md) and writes 4k output tokens. At $3 per million input and $15 per million output, that's $0.15 input plus $0.06 output. About $0.21 per review.

Four reviews a day, thirty days. 120 reviews. $25 burned through your credit.

You're on Pro. The credit is $20. You blow through it in week three. After that, every PR review costs real dollars on extra usage, or your action stops firing. Sonnet 4.6 input runs $3 per million tokens, output $15 per million. Opus 4.8 runs $5 input and $25 output. Cache reads land at 0.10x input, cache writes (5 min) at 1.25x, cache writes (1 hour) at 2.0x.

A heavy Opus headless run can empty a Max 20x $200 credit in roughly a week, per claudefa.st's workload modeling. Frame it as a ceiling, not a guarantee.

Subscriptions used to subsidize agent usage at roughly 15 to 30x compared with API pricing, per Zed's analysis. That subsidy is gone for programmatic calls. Every token your agent burns now counts against the credit, so it is worth working through how to cut Claude Code token costs before the pool goes live.

The Same-Day Model Deprecation

Two model IDs retire on June 15, 2026. The same calendar day the credit pool ships:

After June 15, API requests to the old IDs return errors. Sources: tygartmedia, aiforanything migration guide.

The package names also shift. Anthropic dropped "claude-code" from the SDK package name to underline that the new pool covers any agent built on the SDK, not just Claude Code itself.

Source: theplanettools playbook.

Python users who relied on the inherited Claude Code system prompt have to set it explicitly now. One option object change:

options = ClaudeAgentOptions(
    systemPrompt={"type": "preset", "preset": "claude_code"}
)

That preset hands the SDK the same default system prompt Claude Code used to inject for free.

Three Things Break on June 15

A short list, copied straight from the change log:

1. Calls to claude-sonnet-4-20250514 return errors. Replace with claude-sonnet-4-6-20260217.
2. Calls to claude-opus-4-20250514 return errors. Replace with claude-opus-4-8.
3. The TypeScript package @anthropic-ai/claude-code and the Python package claude-code-sdk are renamed. Install @anthropic-ai/claude-agent-sdk and claude-agent-sdk.

Migration Checklist (Ship Today)

Five steps. The split lands tomorrow, so ship today and monitor the first run.

Step 1. Audit every hardcoded model ID in your repo and your CI. The grep at the top of this post covers it.

Step 2. Bump the IDs to the replacements. Sonnet 4 to Sonnet 4.6. Opus 4 to Opus 4.8. If you are rethinking which model each agent runs, see how to pick the right Claude Code model so the swap also trims spend.

Step 3. Swap SDK packages. In TypeScript:

npm uninstall @anthropic-ai/claude-code
npm install @anthropic-ai/claude-agent-sdk

In Python:

pip uninstall claude-code-sdk
pip install claude-agent-sdk

Then rename ClaudeCodeOptions to ClaudeAgentOptions and add the claude_code system prompt preset if your code expected the old default.

Step 4. Opt in to the Agent SDK credit in your Anthropic account settings. Then enable extra usage with a hard cap. The drain-first plus extra-usage model is what keeps your action from silently going dark mid-month.

Step 5. Deploy today and monitor the first run before the cutover lands. Re-fetch the Anthropic Help Center page to confirm the current numbers. Credit amounts and covered features can shift without warning.

Tools That Make This Easier

Three tools the migration response is short of without:

cc-ledger is a local-first cost tracker that reads Claude Code hooks and writes per-session and per-PR spend to a SQLite file at ~/.cc-ledger/ledger.db. Install:

curl -fsSL https://ccledger.dev/install | bash

After install, every session shows up in the ledger. You see what each PR review actually cost in tokens and dollars. Source: github.com/delta-hq/cc-ledger.

9router is a local proxy that sits in front of Anthropic and falls back through three tiers: your subscription, then cheaper providers, then free ones like Kiro AI, OpenCode Free, and Vertex. RTK token compression saves 20 to 40 percent on top. Endpoint:

http://localhost:20128/v1

Routing production traffic away from official providers carries availability and TOS risk. Treat 9router as a dev-environment safety net, not the main path. Source: github.com/decolua/9router.

Anthropic OpenTelemetry ships built-in spans, metrics, and structured log events for prompts and tool results. If you want a real-time dashboard of where the credit goes, this is the official hook. Docs: code.claude.com/docs/en/agent-sdk/observability.

What If You Do Not Migrate

Two failure modes:

1. The model IDs return errors. Your action goes red. Your cron quits.
2. The SDK package name change kills any npm install or pip install step in CI that pinned the old name.

There is no grace period announced. The deadline is the deadline.

How Build This Now Handles This

If you do not want to babysit Claude Code billing, model migrations, and SDK package renames every quarter, that is what the Build This Now Code Kit handles for you. It is a harness for Claude Code that runs your builds against current Anthropic models. When Anthropic deprecates a model or shifts billing, the kit moves with it, so the model IDs and SDK package names stay current without you tracking a Help Center page. $29 one-time. No subscription.

Frequently Asked Questions

Does Claude Code still work after June 15?

Yes. Interactive Claude Code in your terminal, IDE, and the Claude.ai chat surfaces are unchanged. Only programmatic surfaces (claude -p, GitHub Actions, third-party agents, the Agent SDK) move to the new credit pool.

Will my GitHub Action just start failing on June 15?

It can fail two ways. The model ID retires and the call errors. The package name renames and npm install can't find the old name. Audit both today, before the June 15 cutover.

How fast does the credit drain on Pro?

A typical indie workload of one PR review per push and four pushes per day burns roughly $25 a month at Sonnet 4.6 rates. Pro gives you $20. You hit empty in week three.

Can I share the credit with my team?

No. Per-user only. Not poolable, not transferable, not shareable.

What is the deadline timeline?

Is there a way to keep using the old model IDs?

No. Once the deprecation date hits, requests to the old IDs return errors from the API. You bump the model ID or the call fails.

The June 15, 2026 billing change is a paperwork event, not a product death. Audit your model IDs. Swap the SDK package. Opt in to the credit. Three things on a checklist, and the deadline is tomorrow.