Anthropic June 15 Billing Change: Claude Code Decision Guide

Anthropic announced on May 13, 2026 that effective June 15, 2026, Claude Code and Agent SDK usage move off the shared Claude subscription pool onto a separate monthly credit pool. New pool: Pro $20, Max 5x $100, Max 20x $200, billed at full API rates, no rollover. Here’s the decision table.

Last verified: May 28, 2026.

TL;DR

Your situation	Action
Light Claude Code user, Claude Pro $20	Stay on Pro. $20 credit pool will cover light use.
Medium Claude Code user, Pro $20	Upgrade to Max 5x ($100) or switch to direct API.
Heavy Claude Code user	Direct API billing or Max 20x ($200) with prompt caching.
Team of 3+ heavy users	Anthropic API + console with centralized billing and volume discounts.
You only use Claude.ai chat	No change needed. Chat usage stays on the subscription pool.

What’s actually changing

Before June 15, 2026, Claude Code and Agent SDK programmatic usage shared the same usage pool as your Claude.ai chat sessions. If you hit your subscription cap on Claude Code, your chat experience would also throttle.

After June 15, 2026, there are two separate pools per subscription:

1. Chat pool — Claude.ai web/mobile/desktop, same as today
2. Programmatic credit pool — Claude Code, Agent SDK, any developer-SDK-authenticated usage:
   • Claude Pro ($20/mo) → $20/mo of programmatic credit
   • Claude Max 5x ($100/mo) → $100/mo of programmatic credit
   • Claude Max 20x ($200/mo) → $200/mo of programmatic credit

Key rules:

• Credits are metered at full standard API rates (no subscriber discount)
• Credits do not roll over month-to-month
• Credits are separate from chat — overspending one pool doesn’t affect the other
• If you exceed your credit pool, you can either wait until next billing cycle or switch to direct API billing for that workflow

Why Anthropic did this

Two structural reasons:

1. Unit economics. A small number of heavy Claude Code users were eating most of Anthropic’s compute under the flat subscription. The credit pool caps the per-subscriber liability.
2. Clean separation between consumer and developer products. Claude Pro is now a chat product. Claude Code with serious throughput is now a developer product priced on credits or direct API.

The economic logic is identical to OpenAI’s earlier split between ChatGPT Plus (chat) and direct API access for Codex/programmatic workloads.

What $20 / $100 / $200 of programmatic credit actually buys you

At Claude Opus 4.7 published API rates (~$15/M input, $75/M output tokens, approximate):

Tier	Credit	Rough Opus 4.7 budget
Pro $20	$20	~1-2 hours of intensive Opus coding/mo
Max 5x $100	$100	~5-10 hours of intensive Opus coding/mo
Max 20x $200	$200	~10-20 hours of intensive Opus coding/mo

With prompt caching (which works for Claude Code), real-world hit rates are typically 30-60% on long sessions, which roughly doubles the effective budget. Sonnet 4.6 is cheaper per token and stretches budget further.

Decision tree

Are you a Claude Code user?
├── No → No action needed
└── Yes → How much do you use it per day?
    ├── < 30 min/day → Claude Pro $20 should still work
    ├── 1-3 hours/day → Max 5x $100 OR direct API
    ├── 3-8 hours/day → Max 20x $200 OR direct API
    └── 8+ hours/day or team workload → Direct API + console

Why direct API often wins

For anyone hitting the credit cap, direct API billing is structurally better:

1. No monthly cap. You pay for what you use, no surprise throttling.
2. No rollover loss. Credits expiring monthly is a hidden tax.
3. Prompt caching applies cleanly. Programmatic caching discounts work the same way.
4. Batch API discounts for non-realtime work.
5. Bedrock/Vertex routing. AWS Bedrock and Google Vertex offer the same Claude models with infrastructure benefits — region routing, committed-use discounts, integrated billing with cloud spend.
6. Volume discounts. Available to teams via Anthropic console enterprise tier.

The simplest production pattern:

• Claude Pro $20 for Claude.ai chat surface (good UX, threading, mobile)
• Separate Anthropic API key for Claude Code, billed direct
• For teams: Anthropic console organization with shared API key + cost center attribution

Tools / platforms most affected

These platforms route Claude Code or Agent SDK calls through subscription auth and need migration plans:

• Zed Terminal Threads (Zed 1.3.5, shipped May 20, 2026) — running Claude Code in Zed’s sidebar uses your Anthropic credentials and counts against the new credit pool.
• VS Code with Claude Code extension — same pattern.
• OpenClaw agent runtimes — typically already use direct API keys, so likely unaffected. Verify your installation.
• Custom Agent SDK applications — definitely affected. Move to direct API keys before June 15.

Edge case: enterprise SSO subscribers

If you’re on a Claude Enterprise plan with SSO, the billing change still applies but the credit pools scale by negotiated SKU. Check with your Anthropic account manager — many enterprise contracts already included separate API pools and won’t see meaningful changes.

Three migration patterns

Pattern 1: Solo power user

• Keep Claude Pro $20 for chat.
• Get an Anthropic API key (console.anthropic.com).
• Set ANTHROPIC_API_KEY env var before running Claude Code.
• Now Claude Code charges direct API. Your chat experience is unaffected.

Pattern 2: Small team (3-10 devs)

• Anthropic console organization with shared API key.
• Per-developer cost attribution via project tags.
• Optional: Bedrock or Vertex routing for cleaner cloud billing.
• Each dev keeps a personal Claude Pro for chat if desired.

Pattern 3: Enterprise (100+ devs)

• Anthropic Enterprise contract with negotiated rates.
• Centralized API key management (vault-issued, rotated).
• Detailed usage telemetry per dev / project.
• Optional: deploy Claude via AWS Bedrock or Vertex for unified cloud spend.

What to do this week

1. Audit your usage. Look at the last 30 days of Claude Code activity. If you’re a heavy user, plan now.
2. Get an API key at console.anthropic.com before June 15. Test direct API billing on a small workflow.
3. Decide your pattern (above).
4. Set up cost alerts — Anthropic console supports per-organization spend alerts.
5. Tell your team. If anyone on your team is heavy on Claude Code, they need to know.

Verdict

The June 15 change is a clean economic split between Claude.ai (chat product) and Claude Code (developer product). For light users, the new $20/$100/$200 credit pools are fine. For anyone hitting the cap, direct API billing is structurally better — no rollover loss, prompt caching applies, and you can route through Bedrock or Vertex for cloud-billing integration.

Plan now. The migration is straightforward and the people who do it before June 15 will simply not notice the change.

Sources: Anthropic billing announcement (May 13, 2026), Codersera June 15 billing guide, FindSkill.ai decision table, Anthropic API pricing docs, Verdent Claude Code pricing 2026 guide.