Anthropic ends the subscription subsidy for agentic usage June 15 — Agent SDK and claude -p get a separate credit pool
Chris Harper
2 min read
Jun 13, 2026 · 12:07 UTC
On the same day Claude 4 APIs retire, Anthropic is restructuring how subscription plans cover agentic usage. Starting June 15, the following workloads no longer draw from your regular subscription usage limit:
- Claude Agent SDK (programmatic calls)
claude -p(non-interactive pipeline mode)- Claude Code GitHub Actions
- Third-party apps that authenticate through the Agent SDK
These will instead consume a new Agent SDK Credit pool, billed at full API list prices with no rollover. Monthly allocations:
| Plan | Monthly Agent SDK Credit |
|---|---|
| Pro | $20 |
| Max 5x | $100 |
| Max 20x | $200 |
Interactive Claude usage — Claude.ai chat, Claude Code in the terminal, Claude Cowork — is not affected.
What this costs in practice. Analysis by beam.ai puts the effective price increase for heavy agentic workloads at 12×–175× compared to previous flat-rate bundled access. A $20 Pro credit covers roughly 30–50 medium coding-agent tasks per month at Sonnet 4.6 pricing.
Actions before Monday. First: claim your Agent SDK credit in Claude account settings if you haven't — Anthropic sent claim emails around June 8, and the credit must be activated before it refreshes automatically each cycle. Second: audit your token volume for claude -p runs, CI agents, and GitHub Actions. If you're running multi-step coding agents at meaningful volume, check whether the Pro credit is sufficient or if you need the Max 5x/20x tier. For high-volume pipelines, consider routing cheaper subtasks to Haiku 4.5 to preserve the credit for complex reasoning steps.
Sources: The New Stack: Anthropic splits billing, TechTimes: Anthropic ends subscription subsidy, beam.ai: what AI agents actually cost, ChatForest: billing split guide, DevToolPicks