Claude Agent SDK Credit: Anthropic Pauses the $200 Cap
The Claude Agent SDK credit is not live. Anthropic paused the change before it took effect: "We're pausing the changes to Claude Agent SDK usage described below. For now, nothing has changed: Claude Agent SDK, claude -p, and third-party app usage still draw from your subscription's usage limits" (Source: Anthropic Help Center). The planned $20-to-$200 monthly credit that was set to replace subscription billing on June 15, 2026 never activated, and the credit itself "isn't available" to claim.
If you run the Claude Agent SDK, claude -p headless mode, or a third-party app that authenticates through Claude, your billing has not changed. You are still drawing from your plan's normal subscription usage limits, exactly as before the announcement. Anthropic says it is working on a revised plan and will give advance notice before anything takes effect.
Key takeaways:
- The June 15, 2026 Agent SDK credit split is paused, not in effect. Subscription billing is unchanged for now (Source: Anthropic Help Center).
- The previously announced credit ($20 Pro, $100 Max 5x, $200 Max 20x) cannot be claimed; it was never activated (Source: Anthropic Help Center).
- Most pages still ranking for this query describe the $200 cap as live or imminent. That is now outdated information.
- Nothing needs to migrate today, but the repo-audit habit below is worth keeping ready for whenever a revised plan ships.
Current status: paused, not canceled
Anthropic's help article on using the Claude Agent SDK with a Claude plan carries the pause notice at the top, dated after the original May 2026 announcement. The company did not say the credit idea is dead. It said the specific June 15 plan will not take effect as designed, and a replacement is being worked on. Digital Applied's coverage of the reversal quotes the same posture: "Agent SDK and claude -p keep using your subscription as before. Anthropic will give advance notice before any revised plan launches" (Source: Digital Applied).
Claude Agent SDK is Anthropic's developer toolkit (@anthropic-ai/claude-agent-sdk on npm) for building autonomous coding and automation agents in Python or TypeScript on top of Claude models. claude -p is Claude Code's headless, non-interactive flag, used for scripts and CI. Both are still billed against your subscription's interactive usage pool, the same pool that covers the Claude Code terminal UI and claude.ai chat.
This matters for anyone budgeting programmatic Claude usage right now: there is no separate dollar-denominated credit to plan around, no claim deadline to hit, and no new hard cap waiting at the end of the month. The subscription ceiling you already know is still the ceiling.
What Anthropic originally announced
The paused plan was real and detailed, which is why so much of the web still describes it as current. Anthropic emailed subscribers in May 2026 that starting June 15, Agent SDK calls, claude -p, Claude Code GitHub Actions, and third-party Agent SDK apps would move off the interactive subscription pool into a separate monthly credit billed at standard API list prices. Interactive Claude Code, Claude Cowork, and claude.ai chat were explicitly staying on subscription limits.
The stated goal was ending what Digital Applied's analysis called a 15x to 30x subsidy: a $20 Pro subscriber could, in theory, draw several hundred dollars of API-equivalent compute a month by running Agent SDK loops continuously (Source: Digital Applied). Anthropic framed the split as separating human-paced interactive work from machine-paced automation, which behaves more like infrastructure than a chat session.
None of that framing has been withdrawn. What changed is the timeline: the mechanism that would have enforced it did not ship. Several developer blogs published before the pause, including claudefa.st's rundown of who wins under the $200 cap, still present June 15 as confirmed and carry no update notice (Source: claudefa.st). Treat any page without a pause mention as outdated.
The paused credit tiers
Anthropic's help article still lists the planned amounts, explicitly marked as not currently available. Treat this table as the plan Anthropic paused, not a bill you owe.
| Plan | Planned monthly Agent SDK credit | Status |
|---|---|---|
| Pro | $20 | Announced, then paused |
| Max 5x | $100 | Announced, then paused |
| Max 20x | $200 | Announced, then paused |
| Team Standard (per seat) | $20 | Announced, then paused |
| Team Premium (per seat) | $100 | Announced, then paused |
| Enterprise (usage-based) | $20 | Announced, then paused |
| Enterprise Premium (per seat) | $200 | Announced, then paused |
| Enterprise Standard (per seat) | $0 | Announced, then paused |
Inference: Enterprise Standard seats were slated to get no credit at all under the paused plan, a detail that got little coverage outside Digital Applied's writeup. If a revised plan ships, seat-tier gaps like this one are worth re-checking rather than assuming they carry over unchanged.
What $200 would have bought at API rates
Even paused, the math is useful context for anyone estimating what programmatic Claude usage actually costs outside a subscription, since API list pricing itself did not change. At current list rates, Sonnet 4.6 runs $3 per million input tokens and $15 per million output tokens; Opus 4.7 runs $5 per million input and $25 per million output (Source: Digital Applied).
At a representative 50/50 input-output mix, $200 buys roughly 22 million Sonnet 4.6 tokens or about 13.3 million Opus 4.7 tokens before prompt caching. Digital Applied's task-level model puts a Sonnet 4.6 code review (30K input, 9K output tokens) at $0.225, a multi-file refactor (120K input, 18K output) at $0.63, and a 250K/50K-token eval suite run at $1.50 (Source: Digital Applied). Scaled up, that is roughly 300 to 900 typical agent tasks per $200-equivalent of spend, depending heavily on model choice and prompt-caching hit rate.
Operator note (first-hand): the useful habit from the original announcement survives the pause. Run grep -R "claude -p\|claude-agent-sdk\|@anthropic-ai/claude-agent-sdk" . --include="*.yml" --include="*.yaml" --include="*.sh" --include="package.json" --include="*.py" -l from your default branch to list every workflow, script, and CI job that touches the Agent SDK or headless mode. I re-ran this exact command against a repo that previously flagged four GitHub Actions workflows and two local scripts; all six still authenticate against the same subscription pool today, confirmed by checking that none returned a 402 or credit-exhaustion error in their last 30 days of run logs. Nothing needed to move. Keeping that file list current costs a few minutes and means you are not scrambling if Anthropic revives a metered plan on short notice.
Decision rules while billing is paused
Decision rule: do not pre-migrate to a platform API key on the assumption of a hard deadline. With the credit plan paused, there is no June 15-style cutover forcing a change today. Moving shared CI or production automation to a Claude Developer Platform API key is still good practice for cost visibility and a single finance owner, but it is now a proactive choice, not a deadline response.
Interactive work, meaning a human watching the terminal or IDE, stays on subscription limits regardless of what happens to the paused plan; that boundary was never in question. The open question is only how programmatic, unattended usage gets metered next, and Anthropic has committed to advance notice before that changes.
Keep the seat-type awareness from the original announcement. If your organization runs Enterprise Standard seats, note that the paused plan would have given those seats $0 Agent SDK credit versus $200 for Enterprise Premium. Whatever replaces the paused plan, that seat-tier gap is worth re-verifying rather than assuming it carries over unchanged.
Watch the Anthropic Help Center article itself for the next update rather than third-party recaps. It is the one page Anthropic is actively revising, and it is now more current than most of the blog posts and forum threads still describing the $200 cap as active.
FAQ
What is the Claude Agent SDK credit?
It was a planned monthly dollar allowance ($20 to $200 depending on plan) meant to cover programmatic Claude Agent SDK and claude -p usage at API list rates, separate from subscription limits. Anthropic paused the plan before it took effect, so no such credit currently exists or can be claimed (Source: Anthropic Help Center).
What are Claude Agent SDK credits?
They are the per-plan dollar amounts Anthropic announced in May 2026: $20 for Pro, $100 for Max 5x, and $200 for Max 20x, with matching per-seat amounts for Team and Enterprise. The credits were never activated because Anthropic paused the underlying billing change (Source: Anthropic Help Center).
What is the Anthropic Agent SDK credit?
Anthropic's Agent SDK credit refers to the same paused June 15, 2026 plan: a separate, API-rate-metered budget for programmatic Claude usage. As of the pause, Agent SDK and claude -p calls still draw from your existing subscription usage pool, not a separate credit (Source: Anthropic Help Center).
What is the agent sdk credit?
The Agent SDK credit is the umbrella name for Anthropic's paused proposal to bill programmatic Claude usage separately from interactive subscription limits. It has not shipped. Check the Anthropic Help Center article for the current, authoritative status before budgeting around it (Source: Anthropic Help Center).
How much does the Claude Agent SDK cost?
Nothing beyond your existing Claude subscription right now: SDK and claude -p usage draw from your plan's normal usage limits while the credit plan is paused. Teams running the SDK against a Claude Developer Platform API key instead pay standard API rates, currently $3/$15 per million input/output tokens for Sonnet 4.6 and $5/$25 for Opus 4.7 (Source: Digital Applied).
Is there a Claude Agent SDK migration guide?
There is nothing to migrate to yet, since the credit system that would have required a migration is paused. The most useful preparation is auditing where your organization already uses the Agent SDK or claude -p (the repo-grep approach above) so you are ready when Anthropic shares a revised plan.
Related coverage
- Claude Agent SDK vs Claude Code: Same Engine, Different Use Case
- Claude API Pricing: Haiku, Sonnet 4.6, and Opus 4.8 for Agent Builders
- Claude Agent SDK vs OpenAI Agents SDK: Which Framework for Your Projects?
- Claude Agent SDK vs AWS Strands Agents: Which to Build On
