Anthropic temporarily suspends Agent SDK billing under a new fee policy; subscription subsidies can be as high as 30 times.

Anthropic originally planned to separate the usage of the Claude Agent SDK from subscription plans and switch to API-based billing starting June 15, 2026, but suddenly announced a suspension on the day the new system was to be implemented. The removal of Claude Code from the Pro plan in April 2026 previously triggered a wave of cancellations and was seen as a warning sign for this abrupt halt.
(Background: Anthropic itself triggered the Claude Code cancellation wave, giving OpenAI Codex a major marketing boost)
(Additional context: What is Claude? Costs, features, Claude Code, Cowork — The most comprehensive guide to 2026 Anthropic)

Table of Contents

Toggle

• Original plan: separating Agent usage from subscriptions
• Four examples of when "calling the Agent SDK" occurs
• Key figures: subscription subsidies are 15 to 30 times the API rate
• The "pause": redesign or other considerations

June 15 was originally the date for the rollout of Anthropic’s new Agent SDK billing system. However, on that day, the company suddenly announced a "pause," allowing Claude Agent SDK users to continue with the existing subscription plans that had more generous weekly usage limits. Behind this process is a bill estimated to subsidize API costs by 15 to 30 times per month.

Original plan: separating Agent usage from subscriptions

In mid-May, ClaudeDevs announced the new system, explaining the scope of the adjustment.

The logic of the original plan was to separate the usage of "programmatic calls to Claude" — such as third-party apps, headless commands (claude -p), Claude Code GitHub Actions, and other automated calls to the Claude Agent SDK — from the standard subscription usage pool.

After separation, this portion would be billed at Anthropic’s current API rates, with each paying subscriber receiving a monthly "usage credit" equivalent to their subscription fee (simply put, the amount you pay monthly determines your token quota).

Starting June 15, paid Claude plans can claim a dedicated monthly credit for programmatic usage.

The credit covers usage of:
– Claude Agent SDK
– claude -p
– Claude Code GitHub Actions
– Third-party apps built on the Agent SDK

— ClaudeDevs (@ClaudeDevs) May 13, 2026

Specific numbers: approximately $20 monthly quota for Pro plans, about $100 for Max 5x plans, around $200 for Max 20x plans; enterprise Standard plans are not included in this mechanism. The quota is calculated per user, non-shareable, and cannot be carried over to the next month.

This "non-sharable" design has the greatest impact on teams. Development teams running shared CI/CD pipelines cannot pool their quotas; once the quota is exhausted, automated operations in production must switch to direct API billing. In other words, the new system effectively separates "heavy automation" and "subscription plans" into two different payment models.

Interactive usage, such as Claude Code interactive terminals, Cowork, and claude.ai web chat, remains within the original subscription usage pool and is unaffected. Only programmatic SDK calls are separated.

What are four examples of "calling the Agent SDK"?

1. You use third-party tools like Cursor, Zed, or Cline, which have a "use Claude to modify my code" button. When you click it, it doesn’t go to claude.ai but calls Claude via the Agent SDK. A single click might trigger dozens of consecutive calls.

2. You write a script that batch processes files, e.g., claude -p (headless mode): you issue one command that instructs Claude to translate or correct 200 files in a folder. You run the command and go get coffee; it runs 200 times — all via the Agent SDK.

3. Automated code review teams on GitHub: whenever new code is uploaded, a bot automatically calls Claude to review it (via GitHub Actions). No manual typing, purely event-triggered automation, possibly running hundreds of times a day.

4. Your own automation bot: for example, a script that reads your customer service emails overnight and drafts replies with Claude. While you sleep, it’s working.

Key figures: subscription subsidies are 15 to 30 times the API rate

Why has the developer community reacted so strongly? Because the current subscription plan provides astonishing subsidies for Agent usage.

fazm.ai’s analysis estimates that Claude Opus subscription users only need to send 2-3 messages daily for the subscription to become more cost-effective than using the API; the actual value of the subscription can be several times the monthly fee. Industry estimates suggest that the subsidy for Agent usage via subscription is about 15 to 30 times the API price; once the new system is implemented, heavy Agent SDK users will face costs roughly 5 to 10 times higher, according to Zed Industries’ observations.

In other words, the weekly usage caps in the subscription plan allow heavy users to extract value far exceeding the same price point as API costs. This is a long-term structural loss for Anthropic, but for the developer community, it’s one of Claude’s core competitive advantages over direct API calls.

This contradiction has a precedent in April 2026: Anthropic removed Claude Code from the $20 Pro plan, triggering a wave of cancellations and shifting some traffic to OpenAI’s Codex. That incident was the immediate background for this abrupt halt. Developer Peter Steinberger criticized that Anthropic tends to include popular features in its own products while shrinking the space for external alternatives, a pattern that has accumulated considerable dissatisfaction among developers.

"Pause": redesign or other considerations

Anthropic’s official statement on the 15th was "Nothing changes for now," indicating that they will readjust the plan with the goal of "making it more aligned with how users actually develop with Claude subscriptions," and promising that if they restart the new system, they will notify well in advance.

As for the true reason behind the suspension, outside speculation centers on two points: first, OpenAI is rumored to be planning a significant price cut for its API, sparking a price war, and raising the actual cost for Anthropic’s users at this moment is highly unfavorable; second, analysts speculate that Anthropic has submitted an IPO application, and losing developer users and compressing valuation before going public is not an ideal scenario.

Both of these are currently unconfirmed rumors; Anthropic has not publicly confirmed either.

It’s worth noting that the "pause" itself is not a withdrawal. Anthropic’s wording leaves room for future resumption, only promising to adjust the design and notify in advance. For the developer community, this means the fundamental contradiction in billing — the usage gap between a human user and an autonomous agent — has not disappeared, only temporarily postponed.