Anthropic is splitting Claude's subscription billing in two. Starting June 15, 2026, anything that runs through the Claude Agent SDK or the claude -p command stops drawing from your monthly subscription pool and starts draining a separate, fixed credit instead. Pro gets $20 a month, Max 20x gets $200, and interactive Claude usage stays untouched.
The new arithmetic
Anthropic's support article lays out the breakdown: $20 for Pro and Team Standard seats, $100 for Max 5x and Team Premium seats, $200 for Max 20x and Enterprise seat-based Premium. Usage-based Enterprise accounts get $20. Standard seats on seat-based Enterprise plans get nothing.
The credits are per-user. They don't pool across a team, don't roll over to next month, and they drain before any other source of usage. Once they run out, Agent SDK requests either fall through to extra usage at standard API rates (if enabled) or stop entirely. The subscription limits sit there, reserved for interactive work.
The interesting wrinkle: third-party agents that authenticate through the SDK are now technically welcome again. OpenClaw, OpenCode, anything else that wraps the SDK. They'll burn that same monthly credit.
April, but quieter
On April 4, Anthropic blocked Claude subscription authentication from working with third-party CLIs. The announcement came from Boris Cherny on X with about a day's notice, and the rationale was straightforward enough: subscribers paying $20 or $200 a month were piping their OAuth tokens into OpenClaw-style harnesses and burning what would have cost ten times that on the API.
That was the actual problem. Not third-party tools as a category, but the math. A Max subscription costs $200; an equivalent month of heavy agentic API usage can run past $1,000. VentureBeat's coverage called the gap "compute arbitrage" and said the new credit ends the era. That's about right.
What's $20 actually worth?
That's the question Pro subscribers running automation should be asking. Before April, Agent SDK and claude -p calls drew from the same pool as interactive Claude Code, which meant the same subsidized headroom. After June 15, they don't. A Pro user gets exactly $20 of API-priced agent usage per month. At current Sonnet rates, that buys a meaningful chunk of work. It does not buy what subscription users got six months ago.
And the $20 figure being identical to the Pro subscription itself? Makes the framing feel less like a bonus and more like Anthropic charging twice for the same money. The developer reactions VentureBeat quotes called it a significant reduction in the value of subscriptions. Hard to argue.
The flip side: anyone who only uses Claude interactively will never notice this change. Their limits don't move. The split is targeted at programmatic users, and it's priced like API access because that's effectively what it is.
And the weekly limits
The SDK change isn't landing in isolation. On May 13, Anthropic raised Claude Code's weekly limits by 50% through July 13 for Pro, Max, Team, and seat-based Enterprise users. That came on the heels of a May 6 announcement that doubled five-hour rate limits and stripped out peak-hour throttling for Pro and Max accounts.
All of which traces back to the SpaceX deal for the compute capacity of the Colossus 1 data center in Memphis. 300 megawatts. More than 220,000 NVIDIA GPUs. The company that spent Q1 quietly cutting users off mid-session is now, briefly, in a window of plenty.
You can guess where this goes next. Claude Opus 4.7 launched recently. Bigger models eat more compute. The relief lasts until the next hungry model lands and someone, somewhere, starts complaining about limits again. That's the cycle.
What happens June 15
Pro, Max, Team, and Enterprise subscribers should receive an email by June 8 to claim their credit. It's a one-time opt-in; after that the credit refreshes automatically each billing cycle. Anyone running production automation through the Agent SDK has roughly a month to look at their actual monthly token spend and decide whether the credit covers it.
If you're using an API key through the Claude Developer Platform, none of this applies. Pay-as-you-go continues. No credit, no separation.
The cleaner read on the whole move: Anthropic is converging its subscription and API products. Interactive use stays on subscriptions. Programmatic use, even when it wears a subscription badge, gets priced like the API. The era when a $20 plan could quietly pretend to be a $1,000 one is over.
