Anthropic has put a planned shift in how it bills for its Claude Agent SDK on hold after criticism from developers using the tool. The company had been moving toward token-based billing for the software development kit, but it is now pausing those changes while it reconsiders the approach.
The Claude Agent SDK is part of Anthropic’s growing set of developer tools for building applications around its models. Billing changes tied to tokens, the units of text processed by large language models, can significantly affect costs for software teams. For developers building agentic systems, even small pricing adjustments can change how practical a product is to run at scale.
According to the reporting, the pause comes after the new billing approach drew complaints from some users. The change was intended to make pricing more closely reflect model usage, but developers appeared concerned about the impact on predictability and expense. Anthropic has not said it is abandoning the idea entirely, only that the rollout is being paused for now.
The decision highlights a broader tension in the AI industry as companies try to balance technical flexibility with pricing models that customers can understand. Token-based charging is common across the sector because it ties revenue to usage, but it can also make budgets harder to forecast, especially for products that process large volumes of text or rely on multiple model calls.
For teams using agent frameworks, billing is not just an accounting issue. It can shape product design, determine how often an application invokes a model, and influence whether a project remains economical over time. That makes even modest pricing revisions important for developers, particularly those working on experimentation-heavy products.
Anthropic has been positioning Claude Agent SDK as a way for developers to build and orchestrate AI agents more easily. Any change to the pricing structure for that tool is likely to attract attention from users who depend on stable costs while testing and deploying applications.
The pause suggests Anthropic is taking developer response into account as it evaluates how to price the SDK going forward. The company has not publicly detailed what alternative billing model it might adopt, or whether it will revisit token-based pricing later.
For now, the move leaves current billing expectations in place while Anthropic reassesses its plans. It also adds to the ongoing discussion across the AI sector about how infrastructure, model access, and developer tools should be priced as usage grows and applications become more complex.
Anthropic has not indicated a timeline for any revised policy.