Anthropic Backs Off Its Claude Agent SDK Billing Overhaul on the Day It Was Due — “Nothing Changes for Now”

Anthropic has pulled back a contentious change to how it bills the Claude Agent SDK, scrapping the overhaul on the very day it was due to take effect. In an email to customers, the company said simply: "Nothing changes for now." Usage from the Agent SDK, the claude -p command, Claude Code GitHub Actions and third-party agents will keep drawing from regular Pro and Max subscription limits, exactly as it does today.

The plan Anthropic abandoned, announced weeks earlier and slated for June 15, would have carved that usage out of subscriptions and into a separate monthly credit pool — roughly $20 on Pro and $100 or $200 on the Max tiers — metered at full API rates once exhausted, with no rollover. For developers running agents around the clock, that threatened to convert a predictable flat fee into a usage-based bill that could balloon unpredictably. The reversal lands the same week GitHub's own switch to usage-based billing drew loud complaints from its users.

The backlash was not only about price; it was about Anthropic changing the meaning of a subscription after developers had built workflows on top of it. The company had already barred third-party tools like OpenClaw from subscription limits back in April, and OpenClaw creator Peter Steinberger accused Anthropic of absorbing popular open-source features into its own products and then locking the alternatives out. Pulling agent usage out of subscriptions would have closed the last workaround keeping those tools viable on a paid plan.

Timing made the climbdown almost inevitable. Rival OpenAI is reportedly weighing steep cuts to its API token prices, and shifting customers onto more expensive usage-based billing in the middle of a price war would have been self-defeating. Anthropic is also marching toward an IPO, where customer churn from an unpopular pricing move would dent its valuation — and it is already managing friction from a U.S. order that cut off global access to its Fable 5 and Mythos models for non-U.S. users.

For now, "for now" is the operative phrase. Anthropic has not said the idea is dead — only paused — and the economics that motivated it, chiefly the cost of subsidizing always-on agents on flat-rate plans, have not gone away. Developers get a reprieve, but the question of who pays for autonomous agents at scale is merely deferred, not settled.