Anthropic has paused its plan to introduce token-based billing for the Claude Agent SDK, stepping back from a pricing change that had caused concern among developers building agentic applications on top of the platform. The company confirmed the pause without specifying a new timeline for when, or whether, the per-token billing model would be reintroduced.

What Was Planned and What Changed

Earlier this month, Anthropic announced it would end flat-rate agent access on June 15, moving instead to a consumption-based model that would bill developers for each token used during agentic tasks. The shift was framed as a way to better align costs with actual usage, particularly as agent workflows can consume large numbers of tokens through multi-step reasoning and tool calls. However, the announcement caught many developers off guard, particularly those who had built products assuming a more predictable cost structure.

Key Facts

  • Anthropic has paused token-based billing for the Claude Agent SDK indefinitely
  • The original deadline for the billing change was June 15, 2025
  • The reversal follows developer complaints about cost predictability
  • No new timeline has been provided for reimplementing the pricing model
  • Flat-rate access for existing users appears to remain in place during the pause

The decision to pause reflects a recurring tension in the AI developer tools market: companies need billing models that scale with infrastructure costs, but developers building on top of those platforms need stable, foreseeable pricing to run their own businesses. Agentic workflows are especially sensitive to this dynamic because a single user action can trigger dozens of model calls, making per-token billing hard to cap or predict. The original plan had proposed splitting Agent SDK billing away from standard subscriptions, a structural change that added another layer of complexity for teams managing multiple Anthropic products.

"We're pausing the rollout of token-based billing for the Agent SDK while we work through feedback from developers."Anthropic, via developer communications

Broader Context for Anthropic's Developer Platform

The Agent SDK has become an increasingly important part of how Anthropic positions itself for enterprise and developer adoption. The SDK underpins a range of integrations, including Apple's decision to embed Claude inside Xcode 27 and its Foundation Models framework, which brought the agent tooling to a large new audience of iOS and macOS developers. Any billing instability at the SDK level carries downstream consequences for those partnerships and for the broader ecosystem being built around it.

Anthropic has been moving quickly on multiple fronts this year, from model releases to tooling acquisitions. The company acquired API SDK specialist Stainless to strengthen its developer infrastructure, a move that signaled long-term investment in how developers interact with Claude programmatically. Pausing a billing change mid-rollout is an unusual step, but it suggests the company is prioritizing developer trust over short-term revenue optimization, at least for now.

It is worth noting that token costs are a real constraint for Anthropic as it scales inference across its model family, which now spans several tiers of capability and context length. Finding a billing model that works for both the company and its developer base is not a trivial problem. The pause buys time, but it does not resolve the underlying question of how to price agentic usage fairly.

Developers who had been preparing for the June 15 cutover are advised to monitor Anthropic's official documentation and developer changelog for updates on when a revised billing structure might be announced. The company has not ruled out returning to a consumption-based model for agent tasks, so teams should continue planning for that possibility even as the immediate deadline has been lifted.

Further reading: Learn more about Claude's model family, read our background on Anthropic, or browse the latest Claude AI news.