Claude's Programmatic Billing Split: What You Actually Pay After June 15
Anthropic splits Claude subscriptions June 15—your $20 Pro credit buys fewer agent runs than you think. Here's the math.
We ran the math on Claude Pro’s $20 programmatic credit. Anthropic’s split takes effect June 15, 2026, carving out separate monthly budgets for programmatic use (Agent SDK, Claude Code automation, third-party tools) and billing them at full API rates instead of flat-rate subscription limits. If you run agents daily, you need to understand what that credit actually covers, and what happens when it runs out.
The New Credit Structure
Starting June 15, Claude subscriptions split into two buckets:
- Interactive use (web chat, Claude Code IDE): unchanged. Still draws from your regular subscription limits.
- Programmatic use (Agent SDK,
claude -p, GitHub Actions, third-party agent authentication): now draws from a separate monthly credit, metered at API prices.
Here’s what each tier gets:
| Plan | Monthly Programmatic Credit |
|---|---|
| Pro | $20 |
| Max 5x | $100 |
| Max 20x | $200 |
Credits reset monthly and do not roll over. This is a significant departure from how subscriptions worked before. There’s no buffer, no spillover into your interactive pool. When the credit expires on the 30th or 31st, it’s gone.
The Token Math: How Fast Does $20 Disappear?
To understand what that credit actually buys, we need to price a real agent run. Claude Sonnet 4.6 costs $3 per million input tokens and $15 per million output tokens on the API.
We modeled a realistic agentic workflow, like an automated code review agent or customer-support triage loop that processes a request through multiple turns.
One agent invocation:
- System prompt + context: 8,000 tokens (fixed per run)
- User query: 2,000 tokens
- First Claude response: 1,500 tokens output
- Tool result feedback: 3,000 tokens
- Second Claude response: 1,200 tokens output
Total per run: 15,700 tokens in, 2,700 tokens out
Cost per run at Sonnet 4.6 rates:
- Input: (15,700 / 1,000,000) × $3 = $0.047
- Output: (2,700 / 1,000,000) × $15 = $0.041
- Total: ~$0.088 per run
We calculated that your $20 Pro credit covers approximately 227 such runs per month. If you’re running agents daily, that’s about 7–8 runs per day—tight, but workable for light automation. For Max 5x ($100), you’d get 1,136 runs. For Max 20x ($200), you’re at 2,272 runs.
But real agent workflows are messier. Multi-step tasks, longer context windows, or heavier tool use can push a single invocation to 40,000+ tokens. One senior data scientist noted the monthly allowance “won’t even last a day of serious work”, and that math checks out if you’re spinning up dozens of agents or running continuous monitoring loops.
What Counts, and What Doesn’t
Not all Claude use moves to the new credit pool. Only programmatic access is metered:
Programmatic (uses the credit):
- Claude Agent SDK (Python / TypeScript libraries)
claude -p(command-line non-interactive mode)- Claude Code GitHub Actions integration
- Third-party applications authenticating via Agent SDK (e.g., OpenClaw, Conductor)
Interactive (uses your regular subscription limits):
- Web chat at claude.com
- Claude Code IDE in your terminal or editor
- Desktop / mobile app conversations
This creates a testing loophole. If you’re building an agent, you can test it interactively in Claude Code without touching your programmatic credit. But once you deploy it to production via the SDK, every call eats into the meter.
The Overage Trap: What Happens When $20 Runs Out
When your monthly programmatic credit depletes, additional requests are billed at standard API rates, but only if you’ve enabled extra usage. If you disable extra usage—which is the safe default—requests simply fail after the credit’s exhausted.
The risk is silent. A runaway agent loop or a prompt injection attack could burn through your budget fast, either halting your pipeline mid-month or quietly racking up charges if extra usage is on. There’s no pooled team budget, no grace buffer, no way to dip into your interactive subscription to cover overage. You get $20, $100, or $200, and that’s it.
For context, check how previous Anthropic pricing moves shaped the market. This split is the most direct meter yet on agentic workloads.
Who Gets Hit Hardest
This change primarily affects:
- SDK-heavy developers: Anyone building autonomous agents, batch processors, or scheduled jobs now operates on a tighter monthly budget with no predictability.
- Team workflows: Credits don’t pool across seats, making shared automations awkward and preventing consolidated team budgets.
- Continuous monitors: Apps that make frequent lightweight SDK calls throughout the month will exhaust credit faster than batch-mode users.
- Third-party integrations: Services using OpenClaw, Conductor, or other agent frameworks authenticated via your Claude subscription now consume the same fixed pool.
If you’re already on Max and running serious agentic workloads, the $100 or $200 monthly credit may feel adequate. If you’re on Pro and relying on daily agent runs, you’re entering uncharted territory. You’ll either need to upgrade or switch.
What To Do Before June 15
1. Audit your current Agent SDK usage. If you run agents today via the API, GitHub Actions, or third-party tools, estimate your monthly token consumption and map it to the new credit. Use the support docs to simulate your usage cost.
2. Choose a ceiling: Stay on plan, upgrade, or move to API direct. Pro subscribers with serious SDK workloads should model the cost of upgrading to Max 5x ($100/mo for 1,136 runs) and compare it to API-direct usage (same rates, no subscription overhead). For some, direct API access becomes cheaper than a subscription.
3. Disable extra usage unless you have a hard reason to enable it. This prevents surprise overages, though it means failed requests once the credit runs out.
4. Set a monitoring alert in Anthropic’s console. Once June 15 lands, watch your programmatic usage daily in the first week. If you’re tracking toward zero credit by mid-month, you have 2–3 weeks to switch strategies.
The Bottom Line
Anthropic’s split isn’t a surprise; the company has been signaling API-first pricing for years. But it does force a reckoning: agent use is now visibly expensive. Your $20, $100, or $200 monthly credit is a meter, not a pass. If you’re running agents in production, you’re no longer buying a subscription; you’re buying a consumption cap.
The only question is whether that cap fits your workload.
Sources:
What we don't know is documented at the end of this article. We update when we learn more.