The Claude Agent SDK Credit, Explained in Plain English (June 15, 2026)
You probably got the email from Anthropic on May 13 about a new Agent SDK credit and immediately thought: "Wait, does this affect me?" Short answer: probably not, unless you're running scripts or GitHub Actions against Claude. We've been wiring Claude into client production workflows since Claude 2, so here's the change in language that doesn't assume you live in a terminal. If you only use Claude.ai chat or Claude Code in your terminal, nothing changes for you on June 15, 2026.
Quick Answer:
- Pro plan gets $20, Max 5x gets $100, Max 20x gets $200 in monthly Agent SDK credit, separate from your subscription.
- It only affects programmatic Claude use (Agent SDK,
claude -p, GitHub Actions, third-party agents), Claude.ai chat and Claude Code in your terminal are unchanged. - Credit resets monthly, doesn't roll over, and you have to opt in once after June 15, 2026 to activate it.
What's actually changing on June 15, 2026?
Starting June 15, 2026, Anthropic is splitting Claude paid plans into two separate billing pools. Your existing subscription keeps covering interactive use (Claude.ai chat, Claude Code in your terminal), and a new monthly Agent SDK credit covers programmatic use: $20 for Pro, $100 for Max 5x, $200 for Max 20x. The base plan prices don't go up.
Anthropic announced the change on May 13, 2026. If you've been tracking the May 2026 rate-limit doubling, this is the second half of the story. Same month, different lever.
Starting June 15, 2026, your Claude subscription is split into two pools: one for chatting with Claude, one for letting code chat with Claude on your behalf.
Wait, what is an "SDK," anyway?
Fair question. Most of the news coverage skipped past it. An SDK (Software Development Kit) is a set of tools that lets a program you wrote talk to Claude. Think of it as a remote control instead of pressing buttons on the TV directly. Your code does the typing.
What's an "agent" here? A script that calls Claude in a loop, makes decisions about what to do next, and uses tools, with no human in the chair. Think of a tireless intern that reads issues, drafts replies, and tags teammates while you sleep.
What is claude -p? It's "print mode." You type one question on the command line, Claude answers once, and the session exits. No back-and-forth, no chat history. That one-shot quality is what makes it programmatic: it's designed to be called by other software, not a person.
If you've never opened a terminal or written a script, none of this is about you. Skip to the next section to confirm.
An SDK is a remote control for Claude. Instead of pressing buttons in the chat window, you press buttons from a program you wrote.
Interactive vs programmatic: the only distinction that matters
Anthropic's whole billing split rests on one idea: who is typing the prompt? If a human is in the chair, it's interactive. If a script or service is doing it, it's programmatic. That's the line that decides which pool pays. Here's how the examples sort:
| Type | What it means | Examples | Which pool pays |
|---|---|---|---|
| Interactive | You're in the chair, typing | Claude.ai chat, Claude Code in terminal/IDE, claude.ai/code | Your existing subscription |
| Programmatic | A script or service does the typing | Agent SDK calls, claude -p, Claude Code GitHub Actions, OpenClaw, third-party agents | New Agent SDK credit pool |
Worth noting: even Claude Code hooks, the little scripts that fire on file save or commit, count as programmatic if they call Claude themselves. If a human types the prompt, your subscription pays. If a script types the prompt, the new Agent SDK credit pays.
How much Agent SDK credit do you get?
Here's what each plan gets every month, on top of your existing subscription:
| Plan | Monthly Agent SDK credit | Rollover | Per-seat or pooled |
|---|---|---|---|
| Pro ($20/mo) | $20 | No (resets monthly) | Per-seat |
| Max 5x ($100/mo) | $100 | No | Per-seat |
| Max 20x ($200/mo) | $200 | No | Per-seat |
| Team / Enterprise | Contact Anthropic | Per Anthropic docs | Pooled across seats |
The credit resets on your billing date, doesn't roll over, and shows up as a separate line item, not as extra "messages" on your subscription. So a Pro plan running idle still gets $20 of programmatic credit baked in, whether you use it or not. Numbers confirmed against the Anthropic help-center page.
Does this affect you? A 3-question check
Run through these three quick questions:
- Do you ever run
claude -pfrom your terminal? - Have you set up Claude Code GitHub Actions on a repo?
- Do you use any third-party tool that calls Claude on your behalf, such as OpenClaw, Zed AI features, custom scripts, or automation glue like Zapier-Claude bridges?
If you answered no to all three, your Claude usage doesn't change on June 15. Keep doing what you're doing. If you said yes to any one of them, read the next section, the math actually matters for your wallet.
If you only chat with Claude.ai or use Claude Code in your terminal, nothing changes on June 15. Period.
A real example: how fast does $20 burn?
Say you set up a GitHub Action that triages 20 issues a day. Each run sends Claude about 15,000 input tokens (issue body plus repo context) and gets back about 2,000 output tokens, using Sonnet 4.5. At Sonnet's API rates ($3 per million input tokens, $15 per million output tokens), each run costs roughly $0.075. Twenty runs a day comes to $1.50. Over a 30-day month, that's about $45.
Your $20 Pro credit covers the first 13 days. After that, you either hit extra-usage rates (standard API pricing kicks in) or your Action stops until next month's reset.
claude -p "Summarize this issue in 3 bullets" < issue.mdA one-shot claude -p call. This kind of programmatic use draws from your new Agent SDK credit.
Pro tip: you can swap Sonnet for Haiku to cut burn-rate on tasks where the cheaper model is good enough. We do this on triage-style automations and roughly halve the monthly burn. For a fuller playbook, our reduce LLM API costs guide walks through prompt caching, model routing, and context-trim tricks.
The hidden gotchas (one-time opt-in, runout, and the May 2026 trio)
Three things that are easy to miss in the announcement copy.
1. The one-time opt-in flag
Per Anthropic's help-center page, the credit doesn't activate by itself. You have to toggle it on once in your account settings after June 15, 2026. Until you do, your script calls hit the old behavior, which on June 15 means they may fail or refuse to run. Set a calendar reminder. We've already seen confused subscribers assume their workflow "just works" because Anthropic sent an email.
2. What happens when credit runs out
By default, programmatic calls stop until your next monthly reset. No surprise charges. If you'd rather keep them running, you opt in to extra usage, which bills at standard API rates (Sonnet $3 input / $15 output per million tokens, Opus higher). Nothing bills without you flipping that toggle.
3. How this stacks with the other May 2026 changes
Three Claude changes shipped together this month: (a) the Agent SDK credit (this post), (b) a 10x API rate-limit jump, and (c) Opus 4.7's new tokenizer, which can produce up to 35% more tokens for the same input. The catch: more tokens means the same script burns through your credit faster. Pro and Max users feel all three at once. If you've also built Claude Skills into your stack, factor those calls in too.
What to do before June 15
A 5-step checklist to run this week:
- Identify whether any of your workflows are programmatic (see the 3-question check above).
- Log into your Anthropic account and find the Agent SDK credit toggle once it appears after June 15, 2026.
- Decide if you'll opt in to extra usage or hard-cap at your monthly credit.
- If you're on Team or Enterprise, check the per-seat vs pooled setting with your admin.
- Bookmark the
/usagedashboard so you can watch burn-rate in real time, Anthropic's Agent SDK docs link to it.
That's it. Most people will check these in under 10 minutes.
How Techsy helps
We help teams wire Claude (and the Agent SDK) into real production workflows, sizing the right plan, estimating burn-rate, and building automations that don't blow your monthly credit. If you're staring at this billing change and not sure how to budget your team's Claude usage on June 15, we can map it with you in a 30-minute call. Pairs well with setting up your CLAUDE.md for repeatable, low-burn agent runs. Get a free consultation →
Frequently asked questions
Is the Claude Agent SDK credit a price increase? No. Pro stays $20/month, Max 5x stays $100/month, Max 20x stays $200/month. The Agent SDK credit is added on top, separate from your existing subscription. If you only use Claude.ai chat or Claude Code interactively, you pay the same as before and gain nothing, but lose nothing either.
Does the Agent SDK credit affect Claude Code in the terminal?
No, not when you're using it interactively. Typing into Claude Code in your terminal or IDE still draws from your regular subscription. Only programmatic calls, claude -p, GitHub Actions, scripts that import the Agent SDK, pull from the new credit pool starting June 15, 2026.
What is claude -p and do I use it?
claude -p is "print mode", a one-shot call you make on the command line. You ask Claude something, it answers once, and the session ends. If you've never typed claude -p into a terminal or wired it into a shell script, you're not using it. Most non-developer subscribers never touch it.
Does my Agent SDK credit roll over if I don't use it? No. The monthly credit resets on your billing date and doesn't carry forward. If you used $5 of your Pro $20 credit this month, you don't start next month with $35 — you start with $20 again. Plan your automations around this.
What happens when my Agent SDK credit runs out? By default, programmatic calls stop until your next reset. No surprise bill. If you'd rather keep them running, you opt in to extra usage in your account settings, which charges standard API rates ($3 input / $15 output per million Sonnet tokens, more for Opus). Nothing bills automatically.
Do I need to do anything before June 15, 2026 to keep my workflows running?
Yes, one thing: after June 15, you have to toggle the Agent SDK credit on in your account settings. It's a one-time opt-in. Until you flip it, programmatic calls may fail. If you run a claude -p cron job or a GitHub Action, set a reminder for June 15.
Can I see how much Agent SDK credit I've used?
Yes. Anthropic exposes a /usage dashboard in your account that shows current-period burn against your monthly credit. Bookmark it before you launch anything that could chew through $20 in a weekend. The Agent SDK docs include a link and the field names.
Does this affect Team and Enterprise plans the same way? Not quite. Pro and Max plans get a per-seat credit ($20, $100, $200 respectively). Team and Enterprise plans pool credit across seats and the amounts are negotiated, Anthropic asks you to contact them directly. Check with your account admin before June 15 to see how your org is set up.