Anthropic Ends Agent Subscription Billing June 15: Your Credit Pool Survival Guide
By EndOfCoding
On June 15, 2026, Anthropic ends flat-rate subscription access for Claude Agent SDK, claude -p, Claude Code GitHub Actions, and third-party Agent SDK apps. These will no longer draw from your subscription's unlimited usage. Instead, they draw from a monthly credit pool: $20 for Pro, $100 for Max 5x, $200 for Max 20x — billed at standard API rates. That's 12 days away. Here's exactly what changes and what to do before the meter starts.
What You'll Learn
Which Claude products are affected and which aren't, what the credit pool costs at standard API rates, how to estimate whether your current usage fits in your plan's pool, and 3 specific optimizations you can make before June 15 to stay within budget.
What Changes June 15
Affected (will draw from credit pool starting June 15):
- Claude Code CLI (
claudecommand in terminal) claude -ppipe usage in shell scripts- Claude Code GitHub Actions
- Third-party Agent SDK apps
NOT affected (stays on subscription as before):
- Claude.ai web chat (claude.ai)
- Claude mobile app
- Direct Anthropic API calls you make with your own API key
Credit Pool by Plan
| Plan | Monthly Credit Pool | Rate if You Go Over |
|---|---|---|
| Pro ($20/mo) | $20 in credits | Standard API rates |
| Max 5x ($100/mo) | $100 in credits | Standard API rates |
| Max 20x ($200/mo) | $200 in credits | Standard API rates |
1 credit = $1 of API spend at standard rates.
What Your Sessions Actually Cost
At Claude Sonnet 4.6 pricing ($3/M input tokens, $15/M output tokens):
| Session Type | Approx Tokens | Approx Cost |
|---|---|---|
| Quick Claude Code task (5 min) | 20K in / 5K out | $0.135 |
| Medium session (30 min) | 100K in / 25K out | $0.675 |
| Long session (2+ hours) | 500K in / 150K out | $3.75 |
| Daily ops pipeline (automated) | 200K in / 50K out | $1.35/day → $40.50/mo |
| GitHub Actions CI (per run) | 50K in / 10K out | $0.30/run |
At Claude Opus 4.8 ($15/M input, $75/M output), multiply by ~5x.
Step 1: Audit What You're Running
List every automated workflow that calls Claude:
# Find claude -p calls in scripts
grep -rn 'claude -p\|anthropic\|claude_code' ~/.zshrc ~/.bashrc ~/scripts/ .github/workflows/ 2>/dev/null
For each: how often does it run? What model? Estimated token count?
Step 2: Project Your Monthly Cost
For your Pro plan ($20 credit pool):
- If you have a daily automation at $1.35/day → $40.50/month → 2x your pool
- If you run Claude Code 30 min/day at $0.675/session → $20.25/month → just over your pool
- 1-2 short Claude Code sessions per day → $4-8/month → well within pool
Step 3: Apply 3 Optimizations Before June 15
1. Enable prompt caching on repeated system prompts (90% discount on cached tokens)
# Add cache_control to your long system prompts
messages = [{
'role': 'user',
'content': [{
'type': 'text',
'text': your_system_prompt,
'cache_control': {'type': 'ephemeral'} # Caches for 5 min
}]
}]
2. Route tasks by model — not everything needs Opus
- Quick lookups, summaries, formatting → Haiku 4.5 (10x cheaper than Sonnet)
- Code generation, debugging → Sonnet 4.6
- Complex reasoning, architecture → Opus 4.8 only when needed
3. Reduce CI frequency — GitHub Actions calling Claude on every push may be daily automation you haven't costed. Move Claude-powered checks to scheduled (nightly) or PR-only triggers.
Step 4: Upgrade Your Plan If Needed
If your projected monthly cost exceeds your pool:
- Pro ($20 pool) → Max 5x ($100 pool): $80 more per month
- Max 5x ($100 pool) → Max 20x ($200 pool): $100 more per month
- Over $200/month: direct Anthropic API with your own billing (pay-as-you-go, no pool cap)
The Anthropic API direct option remains available and unchanged — it's the subscription subsidy that's ending, not API access.
Common Challenges
I don't know how many tokens my Claude Code sessions use: Use the session cost estimator above. A typical 30-minute coding session with Claude Sonnet 4.6 runs $0.50-$1.00. I have automated daily scripts calling Claude: These are the highest-risk workflows. Estimate the daily token count, multiply by 30, compare to your pool. If the math doesn't work, either add caching or upgrade your plan. What if I go over my credit pool: You'll be charged at standard API rates for the overage — the same rates in the table above. There's no hard cap that cuts off access.
Advanced Tips
For power users running daily orchestration pipelines (like the Veridux daily ops pipeline), the Max 20x plan at $200/month credit pool is likely the minimum viable tier. At Sonnet pricing, $200 supports roughly 66M input tokens + 13M output tokens per month — generous for most pipelines. For teams: the credit pool applies per user/subscription. If multiple developers are running Claude Code sessions under a shared plan, the pool may not scale. Direct API keys with your own billing is the right call for team environments.
Conclusion
June 15 is 12 days away. The developers who won't notice the change are those using Claude primarily through the web app or running short, infrequent Claude Code sessions. The developers who will notice are those with automated pipelines, GitHub Actions, or daily heavy Claude Code usage. Run the audit above today. If your projected usage exceeds your pool, apply the optimizations or upgrade before June 15. For the complete prompt library for managing AI coding tool costs — including the Anthropic Credit Pool Budget Planner — see Chapter 17, Prompt 17.319 in the Vibe Coding Ebook at vibecodingebook.com/reader#ch17. And for the latest AI coding tool comparison including cost-per-workflow data, see endofcoding.com.