Anthropic moves Agent SDK and claude -p to a separate credit pool June 15

Anthropic is splitting how it bills Claude usage, and the change takes effect Monday, June 15, 2026. Announced in mid-May through the company's help center, the update separates programmatic usage from the interactive usage that most subscribers know.

Until now, automated workloads have shared the same rate-limit pool as chatting with Claude in the terminal or on the web. That included the Claude Agent SDK, the headless claude -p command, Claude Code GitHub Actions, and third-party agent apps that authenticate through the SDK. Those workloads were effectively subsidized well below standard API list prices.

What changes

From June 15, Anthropic states that "Claude Agent SDK and claude -p usage no longer counts toward your Claude plan's usage limits." Instead, that traffic draws from a new, separate monthly dollar credit, metered at standard API rates with no rollover. According to reporting from The New Stack and several developer breakdowns, the credits are roughly:

• Pro: $20/month
• Max 5x: $100/month
• Max 20x: $200/month
• Team Standard / Premium: $20 / $100 per seat

The credit is granted per user, does not carry over month to month, and several write-ups note you may need to claim it once before the cutover.

What stays the same

The interactive experience is unaffected. Chatting on Claude.ai (web, desktop, mobile), running Claude Code interactively in your terminal, and Claude Cowork all keep drawing from your normal subscription limits exactly as before. Developers who already pay with a direct Anthropic API key also see no change — they were already on metered API pricing.

In short: typing prompts into Claude Code yourself is untouched. Wiring Claude into a script, a CI job, a cron task, or a third-party agent now spends the new credit, and once it's exhausted you pay standard API rates (or the job stops, depending on your setup).

Why it matters for people learning to build

If you're following along in tutorials and using Claude Code by hand, your day-to-day cost model doesn't move. The group most affected is anyone running automated or background agents on a subscription — fleets of headless agents, GitHub Action reviewers, or hobby projects that quietly looped claude -p. Coverage from TechTimes and a detailed user guide frame it as the end of an uncapped subsidy: for heavy programmatic use, the effective price can rise sharply versus the old flat-rate arrangement.

The practical takeaway is to know which bucket your usage falls into. If you automate anything with Claude, check your plan's new credit, estimate your monthly token spend at API rates, and decide whether a dedicated API key or a higher tier makes more sense. The change is also a useful reminder of a broader 2026 trend: as agents get cheaper to run, providers are getting more precise about who pays for the compute when a model works unattended.