Anthropic was set to roll out a new billing model for Claude Agent SDK on June 15, 2026, then stopped it on the same day. The company said that, for now, nothing changes. Users on existing subscription plans can keep using the current weekly limits instead of moving to the new structure immediately.
Planned change would separate programmatic agent usage from subscriptions
Under the original plan, programmatic calls to Claude Agent SDK would be carved out of the normal subscription usage pool. That covered third-party apps, claude -p headless mode, and Claude Code GitHub Actions. Once separated, those calls would be billed at Anthropic’s current API rates rather than through the standard subscription bucket.
Subscribers would receive monthly usage credit equal to the price of their plan. The source lists about $20 for Pro, about $100 for Max 5x, and about $200 for Max 20x. Enterprise Standard was not included in this system. Credits would be calculated per user, with no sharing across teammates and no rollover into the next month. That detail mattered for teams running shared CI/CD pipelines, because once a user’s credit ran out, the workflow would need to switch to direct API billing.
Four common cases that count as Agent SDK usage
The source outlines several examples. One is when a developer clicks a Claude-powered code action inside tools such as Cursor, Zed, or Cline. A single click may trigger many backend requests through Agent SDK rather than a direct claude.ai session. Another is batch work through claude -p, where one command can launch repeated processing across a large number of files.
Two more cases involve automation. GitHub Actions can run code review tasks automatically whenever new code is pushed, and custom bots can call Claude in the background for jobs such as drafting email replies. These uses can stack up quickly because they are event-driven or scheduled, not tied to manual chat sessions. Interactive usage, including Claude Code terminal interaction, Cowork, and claude.ai web conversations, was supposed to remain inside the original subscription pool.
Why developers focused on the subsidy gap
The backlash centered on cost. According to analysis cited from fazm.ai, a Claude Opus subscriber may already come out ahead versus API pricing with only about 2 to 3 messages per day. The broader industry estimate in the source puts the effective subscription subsidy for agent usage at roughly 15x to 30x compared with API rates.
Zed Industries offered another figure: heavy Agent SDK users could see their real costs rise by about 5x to 10x if the new model took effect. That helps explain the intensity of the reaction. For Anthropic, the current arrangement can look like a structural drain. For developers who rely on automation, the same pricing gap has been one of the clearest reasons to stay on subscription plans.
April’s Claude Code change set the stage
This was not the first pricing-related clash. In April 2026, Anthropic removed Claude Code from the $20 Pro plan, a move that triggered cancellations and pushed some usage toward OpenAI’s Codex, according to the source. That episode became an immediate backdrop for the June reversal.
Developer Peter Steinberger, also cited in the source, argued that Anthropic has folded popular functions into its own products while narrowing the room for external alternatives. In this dispute, the concern went beyond price alone. Developers were also reacting to a structure that would split subscription-based work from automated agent-heavy workflows.
Paused, not withdrawn
Anthropic said it would redesign the plan so it better matches how users actually build with Claude subscriptions, and it promised advance notice if it decides to relaunch the billing change later. The company did not say the policy was canceled. It said it was paused.
The source mentions two outside theories for the timing: possible API price cuts from OpenAI, and speculation that Anthropic has filed in preparation for an IPO. Neither point has been publicly confirmed by Anthropic. What is clear now is simpler. The billing change has been delayed, while the underlying tension between human subscription usage and high-frequency autonomous agent usage remains unresolved.