Anthropic moved today to charge OpenClaw users a new fee, separate from their existing Claude subscription. Starting at noon Pacific on April 4, Claude Code subscribers can no longer apply their plan limits to third-party agent harnesses including OpenClaw. Usage will now be billed through a pay-as-you-go system Anthropic calls "extra usage" bundles, or not at all.
The move is distinct from the general April 4 cutoff we covered earlier — that was about session length and API rate limits. This is an integration-level change: OpenClaw has been running on the same billing infrastructure as Claude Code itself, and Anthropic is now carving it out as a separate line item. The practical effect is the same, but the mechanism matters. This is not a limit. It is a repricing.
Peter Steinberger, who created OpenClaw and recently left for OpenAI, said on X that he and OpenClaw board member Dave Morin tried to talk Anthropic out of the change. The best they got was a one-week delay. "Funny how timings match up," he wrote. "First they copy some popular features into their closed harness, then they lock out open source."
The features he is referring to are real. Anthropic shipped MCP (Model Context Protocol), the tool-calling standard, in Claude Code in recent months, a capability OpenClaw helped popularize in the agent framework ecosystem. Claude Code Channels, which lets external services message agents, arrived around the same time OpenClaw was gaining traction on persistent autonomous workflows. These were the capabilities that made OpenClaw useful for more than toy demos.
Boris Cherny, Anthropic's head of Claude Code, disputed the framing on X. The Claude Code team are "big fans of open source," he said, and he personally submitted pull requests to improve OpenClaw's prompt cache efficiency. He framed the change as a capacity decision: Anthropic's subscription infrastructure was not designed to absorb the usage patterns of autonomous agents running continuously across sessions. A single OpenClaw agent running for a day can consume tokens at a rate Aakash Gupta estimated at $1,000 to $5,000 in API costs. Claude's subscription tiers were not built for that.
That cost argument appears technically sound. OpenClaw users running autonomous task loops do generate significantly more token volume than typical Claude Code sessions, which involve direct human prompting. The five-hour session limits Anthropic introduced earlier this year were an earlier attempt to manage the load. The billing change is the more permanent solution.
Anthropic is offering existing subscribers a one-time credit equal to their monthly plan price, redeemable until April 17, and is offering up to 30 percent off extra usage bundles for users who pre-purchase. None of this changes the core math: what was covered by a $20 Pro subscription now costs by the token, and for a heavy user that is a very different number.
The uncomfortable read of the timing is not hard to construct. OpenClaw users were Anthropic's most intensive subscribers. Cutting them off from subscription billing simultaneously trims load on Anthropic's infrastructure, directs heavy users toward API pricing that is more directly profitable, and removes a reason to use a third-party framework that does not feed directly into Claude Code's feature set. The billing change reads as consistent with both a cost management decision and a competitive move. Cherny is not wrong about the token volume. Both things can be true.
OpenClaw continues to work with Claude via API key access and via the new extra usage bundles. It is not dead. But the experience that made it feel like a first-party feature, integrated into the same billing system as Claude Code itself, ended today. For a tool that built its reputation on being the most elegant way to run Claude autonomously, that is a structural change in the relationship between the framework and the model it runs on.
