Anthropic Ends Flat-Rate Claude Access for Third-Party AI Tools

Anthropic, the company behind the popular AI assistant Claude, announced a major change to its subscription model. Starting April 5, 2026, Claude subscriptions will no longer cover usage through third-party tools like OpenClaw. Users who rely on these external platforms will need to pay separately through a usage-based billing system or connect with an API key instead.

What Is Changing and Why

Claude subscriptions currently allow users to interact with the assistant through both Anthropic’s own products and third-party tools known as harnesses. These harnesses connect Claude to external coding environments and automation workflows, letting developers build powerful systems on top of the model. However, Anthropic says the heavy, continuous usage generated by these tools has become unsustainable under flat-rate pricing.

Boris Cherny, head of Claude Code at Anthropic, explained that subscriptions were never designed for the demand patterns that AI agents create. An AI agent is a system that can carry out multi-step tasks on its own, often running for hours without pause. That kind of always-on usage quickly exceeds what a fixed monthly fee can support. Anthropic now wants to prioritize capacity for its direct products and API customers going forward.

How Anthropic Is Softening the Blow

Affected users will not lose access entirely. They can still power third-party tools with Claude through a pay-as-you-go system or by purchasing discounted usage bundles at savings of up to 30 percent. Anthropic is also offering a one-time credit equal to each user’s monthly subscription cost, redeemable by April 17, and full refunds are available via email for those who prefer to cancel.

OpenClaw creator Peter Steinberger called the move sad for the ecosystem but acknowledged Anthropic’s mitigation efforts. He also released efficiency updates to help users cut their API costs. The broader signal is clear: flat-rate pricing may not survive the age of AI-driven automation, where autonomous tools can consume far more compute than any human user.

The decision highlights a growing challenge across the tech industry. As AI agents become more capable and widely adopted, companies building large language models — the deep learning systems that power tools like Claude — will need to find pricing models that balance accessibility with the real cost of running these powerful systems.