Claude's June 15 billing change: programmatic usage moves to a separate credit pool
In brief
From June 15, 2026, Agent SDK, `claude -p`, Claude Code GitHub Actions, and third-party agent apps stop drawing from your subscription's usage limits and start drawing from a separate monthly dollar credit, metered at API list prices. Interactive Claude Code and web chat are unchanged. Here's exactly what changes, the credit per plan, and what to do before the deadline.
Contents
Anthropic announced on May 14, 2026 that programmatic Claude usage is splitting off from subscription rate limits. The change takes effect June 15, 2026. If you run agents, scripts, or CI against your Claude Pro, Max, or Team subscription, this changes your bill and possibly your behavior — so it's worth understanding before the deadline rather than after.
What's changing
Today, programmatic Claude usage draws from the same pool as your interactive chat — your subscription's rate limits cover both. From June 15, programmatic usage moves to a separate, dollar-denominated monthly credit, metered at standard API list prices.
What counts as "programmatic" (moves to the new credit pool):
- The Claude Agent SDK running in your scripts and projects
claude -p— the headless / non-interactive mode of Claude Code- Claude Code GitHub Actions
- Third-party apps that authenticate using Agent SDK credentials (this is the path that lets tools like OpenClaw and editor integrations run on your subscription)
What stays exactly the same (still covered by your normal subscription limits):
- Interactive Claude Code in your terminal
- Claude web and desktop chat
The simple rule: if a human is sitting there typing, it's unchanged. If code is calling Claude on its own, it now spends from the credit pool.
The numbers
Each plan gets a fixed monthly credit, billed at full API list rates:
| Plan | Monthly programmatic credit |
|---|---|
| Pro | $20 |
| Max 5x | $100 |
| Max 20x | $200 |
| Team Standard | $20 / seat |
| Team Premium | $100 / seat |
Three things to internalize about the credit:
- It's per user, not pooled. A 10-seat Team Standard plan does not get $200 in one shared bucket — each person gets their own $20.
- It resets monthly with no rollover. Unused credit disappears at the end of the cycle.
- It's metered at API list prices, not at some discounted subscription rate. $20 of credit buys $20 of API usage — which, with Opus, is not a lot of agent runs.
Why Anthropic is doing this
The honest framing: this ends a period of uncapped, subsidized programmatic access. A subscription was a flat monthly fee, and some users were running heavy automated workloads — agents looping for hours, CI pipelines calling Claude on every commit — that cost far more in compute than the subscription brought in. Moving programmatic usage to metered credit aligns the price with the cost. For light programmatic users the included credit may cover everything; for heavy ones it's effectively a large price increase, because past that credit you pay API rates per token.
What to do before June 15
- Watch for the claim email (around June 8). Anthropic is sending an email to claim your credits. Don't ignore it.
- Audit your real programmatic spend at API rates. Before the change lands, get a number: how much would your current Agent SDK /
claude -p/ GitHub Actions usage cost at API list prices per month? If it's under your plan's credit, you're fine. If it's over, you have a decision to make. The/usagecommand and the Usage & Cost API help you measure this. - Decide your overflow behavior. There's a "usage credits" toggle that controls what happens when the credit runs out — either stop, or keep going and bill at API rates. Pick deliberately; the default could quietly run up a bill or quietly break a pipeline.
- Right-size, or move heavy workloads to the API directly. If you're well over the credit, a higher plan tier, or a proper API key with its own billing and rate limits, may be cleaner than living on overflow.
- Brief whoever owns your CI. GitHub Actions calling
claude -pon every PR is exactly the kind of workload that will eat the credit fast. The person who set that up needs to know before June 15, not after the first surprised invoice.
Who this hits hardest
If you're an Agent Operator running production agents off a subscription, or a developer with Claude wired into CI, this is a real operational change — model it now. If you only ever use Claude by typing in the terminal or the web app, you can stop reading: nothing about your usage changes.
Try this today
Open a terminal and run claude /usage, or pull last month's programmatic usage from the Usage & Cost API, and write down one number: your monthly programmatic spend at API list prices. Compare it to your plan's credit from the table above. That single comparison tells you whether June 15 is a non-event or a thing you need to act on this week.