Claude Agent SDK Gets Separate Billing on June 15 — D-28

EDIBLOG · AI · ANTHROPIC

Claude Agent SDK gets separate billing on June 15 — D-28

Bottom line first. Even with Pro $20 or Max $200, your automation jobs get billed separately. The Claude Agent SDK billing split Anthropic announced back in April is now D-28 away. If you run cron jobs with claude -p, it’s time to check ahead.

EFFECTIVE · D-28 2026/06/15 Pro $20 / Max5x $100 Max20x $200 claude -p · SDK · GHA

01 · BEFORE → AFTEROne pool splits into two

Through 6/14 · single pool
UNIFIED
Interactive + automation, same pool
claude.ai chat same pool
Claude Code interactive same pool
claude -p scripts same pool
Agent SDK jobs same pool
2026 / 06 / 15
From 6/15 · two pools
SPLIT
Interactive / programmatic split
claude.ai chat subscription pool
Claude Code interactive subscription pool
claude -p scripts separate pool
Agent SDK jobs separate pool

02 · CREDIT BY PLANAutomation pool amount by plan

PRO
monthly extra pool
$20
Same as subscription
MAX 5×
monthly extra pool
$100
Same as subscription
MAX 20×
monthly extra pool
$200
Same as subscription
Once depleted
regular pool
No deduction
API-rate billing via extra-usage toggle

The monthly pool expires at month’s end if unused, and it won’t fall back to your regular subscription pool either. If you need more, you have to flip on the “extra usage” toggle to get billed automatically at standard API rates — leave it off and requests just stop. Anthropic announced this change on May 13 via its official @ClaudeDevs account on X, and the next day InfoWorld, The Register, Axios, and VentureBeat all ran explainers. It’s essentially April’s third-party agent block policy reopened in the form of a credit pool.

03 · AFFECTEDUsage deducted from the new pool

▮ Hits the separate pool from 6/15
  • Claude Agent SDK callsAgents built with the Python·TypeScript SDK
  • claude -p (programmatic CLI)Calls from cron / launchd / shell scripts
  • Claude Code GitHub ActionsAuto-review · auto-PR called from CI jobs
  • Third-party tools built on the Agent SDKExternal harnesses like OpenClaw share the same pool

04 · UNAFFECTEDStill deducted from the existing subscription pool

▮ Interactive usage that stays the same
  • claude.ai web · app chatConversations a human types directly
  • Claude Code interactive sessionsSessions a human launches in the terminal
  • Claude CoworkTeam collaboration interface usage

In other words, the pool for “what a human clicks” and “what a script clicks” gets split apart. Call the same Opus 4.7, and which pool you draw from depends on who’s making the call.

05 · FOR DEVSWhat actually changes for you

▮ OUTLINE · the objective rundown 2026 / 05 / 18 · D-28

The era of “one subscription runs everything, automation included” ends on June 14. That said, for everyday users almost nothing changes. It’s just a good moment to check where your own usage actually falls.

Heavily affected

Anyone running claude -p automation on cron, Claude Code Actions inside CI, or an Agent SDK side project.

Barely affected

If you only use claude.ai chat and Claude Code interactive in the terminal. The separate pool just expires unused.

“Even when you call the same model at the same price, we’re entering an era where who makes the call decides which pool you spend.

— Eddie · noted at D-28

If you’re running claude -p, GitHub Actions, or the Agent SDK, it’s worth a usage check before June 14. The subscription pool and the automation pool are separate from 6/15 on.

Disclaimer: Summarized from the official announcement and primary reporting. Confirm exact policy on Anthropic’s official page as of the effective date. No ads · affiliates.


This post is part of an English mirror of a Korean dev experiments blog. Cross-posted for the global developer audience.

Leave a Comment