Anthropic is splitting Claude Code’s programmatic surface off the subscription billing line. Starting June 15, 2026, the Agent SDK, claude -p (non-interactive mode), and Claude Code GitHub Actions stop drawing from Pro and Max usage allowances and start drawing from a separate API credit pool (ClaudeDevs announcement).
What each tier gets as the replacement credit:
- Pro ($20/mo): $20 in monthly API credit for programmatic usage
- Max 5x ($100/mo): $100 in monthly API credit
- Max 20x ($200/mo): larger credit pool, scaled to plan (per Medium summary)
- Interactive
claudeCLI and the Claude Code IDE extensions: unchanged, still inside the subscription
The practical read: if you only use Claude Code interactively in the terminal or IDE, nothing changes. If you’ve been wiring claude -p into shell scripts, cron jobs, CI, or building agents on the SDK, you are now metered at API rates once the credit runs out. A Pro user burning through SDK calls in a CI pipeline will hit the $20 ceiling fast — Sonnet output alone is $15/million tokens.
This is the cleanest move in a rough six weeks for Claude Code billing, which included briefly removing Claude Code from Pro entirely and blocking third-party tools like OpenClaw from subscription usage. Separating programmatic from interactive billing is defensible — automated agents can consume orders of magnitude more tokens than a human typing prompts — but it does quietly reprice the heaviest users.
NOTE
The credit is denominated in dollars at API rates, not in tokens or requests. That means your effective allowance shifts with model choice — Haiku-tier work stretches a Pro credit much further than Opus calls.
If you run Claude Code in CI or scripts, audit your token volume before June 15. Anyone past roughly 1.3M Sonnet output tokens a month on Pro will need to top up with pay-as-you-go API billing or move to Max.