Subscribers woke up to an unexpected email on June 15. Anthropic's planned billing split, which would have moved Agent SDK usage, claude -p, and third-party Claude apps onto a separate monthly credit pool, was paused on the exact morning it was scheduled to take effect. The subject line: "We're pausing the Agent SDK credit change."
No elaborate explanation. Just a single, direct sentence from Anthropic:
"We're not making this change today. We're working to update the plan to better support how users build with Claude subscriptions."
Anthropic, email to subscribers, June 15, 2026
Three things you need to know right now.
- Your current setup still works. Agent SDK,
claude -p, and third-party Claude apps remain on your existing subscription, exactly as before. - The credit pool does not exist. The announced $20/month (Pro) or $100/month (Max) allocation at API token rates is off the table, for now.
- You will hear about changes in advance. Anthropic committed to giving prior notice before anything changes.
Why did Anthropic reverse course?
The company did not say. But the context is hard to miss. Anthropic announced the billing split on May 13, gave users a month, and then reversed on day one.
Here's the thing: not one of Anthropic's primary competitors charges separately for automated usage. GitHub Copilot, Cursor, and OpenAI Codex all bundle agent and programmatic usage into a flat subscription. Singling out SDK usage for extra billing, right as that pricing gap was becoming a talking point among developers, was a risky move. Think of it like a gym that starts charging extra every time you use the treadmill during peak hours, after you signed up for flat-rate access.
| Tool | Automated usage billed separately? |
|---|---|
| GitHub Copilot | No |
| Cursor | No |
| OpenAI Codex | No |
| Claude (paused plan) | No, for now |
Worth noting: The Decoder reported that the reversal likely connects to a brewing price war with OpenAI and broader pressure ahead of a potential Anthropic IPO. That pressure applies across every developer market globally. Claude Code alone generates an estimated $2.5 billion in annualized revenue, and a large share of that comes from automated, programmatic use. Making exactly that usage more expensive, while competitors hold flat-rate pricing, is a hard case to make publicly.
Is the plan dead or delayed?
"We're working to update the plan" reads more like a redesign than a cancellation. The mechanics of separating interactive from programmatic usage are not new: AWS, Azure, and Google Cloud all tier their AI API pricing by workload type. What Anthropic clearly learned is that subscribers expect flat-rate access, and changing that model mid-subscription requires considerably more notice than 30 days.
How an updated plan will look, and when, remains unknown. The one thing Anthropic committed to: advance notice before anything changes.
What should you do right now?
- Nothing needs to change today. Your automated workflows run exactly as they did before June 15.
- Ditch any budget planning based on the old credit model. The $20/$100 pool does not exist. If you built cost projections around it, scrap them.
- Do not assume this pricing is permanent. AI pricing in 2026 is moving fast. If you run significant automated Claude workflows, tracking your token usage now is a smart move. Our generative AI statistics page keeps an updated overview of current API token rates across providers.
