Anthropic "pauses" token-based billing for its Claude Agent SDK

Anthropic recently announced a pause on its planned token-based billing changes for the Claude Agent SDK. These changes, initially set to take effect, would have substantially increased costs for many third-party applications and heavy users of the automation-focused SDK. The decision allows users to continue operating under the more generous usage limits of their existing Claude subscriptions.

The original plan, unveiled in May, aimed to bill Claude Agent SDK usage separately from standard Claude usage via the chat interface or official CLI. This meant that external SDK usage would be charged at Anthropic's prevailing API rates, with subscribers receiving a monthly usage credit equivalent to their subscription price. This represented a significant departure from the previous model, where Agent SDK use was only capped by the weekly limits of a user's current Claude subscription tier, often allowing power users to extract considerably more value than through direct API fees.

Developers had expressed concern, with one analysis suggesting that Claude Opus users could exceed breakeven within the first week under the new pricing. The developers behind the code editor Zed also warned their users of a "major cost increase." The sudden retraction of these pricing changes follows similar token-based billing adjustments by GitHub Copilot, which caused "sticker shock" for many users. The move also occurs as Anthropic prepares for a potential initial public offering.

While this temporary reprieve is welcomed by Claude Agent SDK users, it's likely a short-term solution. Anthropic's Head of Claude Code, Boris Cherny, stated in April that "our subscriptions weren’t built for the usage patterns of these third-party tools," indicating a need to manage capacity thoughtfully and prioritize customers using their products and API sustainably long-term.