Claude Agent SDK Billing Splits Today: When the Subscription Credit Covers Your Agent and When to Route via API Key

Starting today, June 15, 2026, how Anthropic meters Claude Agent SDK usage has fundamentally changed — and if you route production automation through a subscription plan rather than a direct API key, you need to know exactly where the new ceiling sits.

The change is straightforward in policy: Agent SDK usage and claude -p invocations no longer count against your Claude plan's interactive usage limits. Instead, they draw from a new, separate monthly Agent SDK credit. When that credit runs out, additional usage spills to "usage credits" (pay-as-you-go at standard API rates) — but only if you've explicitly enabled usage credits on your account. If you haven't, agent requests simply stop until the credit refreshes next billing cycle.

What the monthly credit actually buys

The credit is per-user and non-pooled. It refreshes monthly with your billing cycle and unused amounts do not roll over.

Team admins: credits belong to individual users. There is no pooling mechanism — you cannot consolidate ten Pro users' $20 credits into $200 for one shared automation account.

At standard Fable-class token rates, a $20 monthly credit covers roughly 400K output tokens (Fable 5 at ~$50/M output). For a lightweight personal automation that fires a few times a day, that's probably enough. For any shared production agent running multiple times an hour, it is not.

The routing decision matrix

This billing split creates a hard fork in how teams should route their agent workloads:

Subscription credit lane — right for:

• Individual developer experimentation and local agent runs
• Low-frequency personal automation (daily summaries, PR reviews on quiet repos)
• Claude Code and Claude Cowork interactive sessions (these stay on the existing subscription meter, unaffected)
• Teams where each developer runs their own agent under their own account

Direct API key lane — right for:

• Shared production automation where a single agent account serves multiple team members
• 24/7 agents or high-frequency polling loops that will exhaust a $20–$200 credit in days
• Teams that need predictable billing with a unified cost center, not per-user credit fragmentation
• Any workflow that cannot tolerate silent request termination when a credit pool runs dry

Anthropic's own support doc is explicit: "Teams running shared production automation should use Claude Platform with an API key for predictable pay-as-you-go billing."

Why this matters for Claude Agent SDK billing governance

The non-pooled, per-user structure introduces a new failure mode for operators who aren't paying attention: a junior developer's agent quietly drains their $20 credit on the first day of the month, then requests stop — with no team-level budget control and no pooling to borrow from.

For teams using a gateway to route agent traffic, this signals the moment to move shared production agents off subscription-backed accounts and onto an organization-level API key with proper rate-limit and cost-control policies. That means:

1. Audit which agents run under subscription accounts vs. dedicated API keys today.
2. Route shared or high-frequency agents to an API key path before they silently exhaust credits.
3. Enable usage credits on any subscription account that must not hard-stop mid-cycle — understanding that this means unbounded spend beyond the monthly credit.
4. Set spend alerts at the platform level if you opt into usage credits, since there is no native cap.

What the credit does not cover

The Agent SDK credit does not apply to:

• Interactive Claude Code sessions (those remain on the subscription usage limit)
• Claude Cowork usage
• API requests made with an API key — those always go to the standard API billing path

This means mixed-use accounts (developers who both interactively use Claude Code and run automated agents) now have two separate meters to monitor.

What to watch

Anthropic has not announced a credit pooling feature for teams. The current architecture explicitly routes shared production automation to the API key path rather than trying to aggregate per-user credits. Watch for:

• Whether Team and Enterprise plans gain an admin-level credit-pooling control
• Whether usage credits can be capped (currently they cannot — enabling them means unbounded spillover)
• Whether claude -p in CI/CD contexts counts against the credit the same way SDK calls do (currently yes, per the support doc)

For teams routing Claude agents through a central gateway like TheRouter — where API keys are managed centrally and cost attribution is tracked per team — the API key path remains the right architecture for anything beyond individual developer experimentation. The subscription credit lane is a convenience feature for personal agents, not a foundation for production routing policy.

Read more on what the Fable 5 model changes mean for your token budget migration and the export control suspension that took Fable 5 and Mythos 5 offline for international access.