Anthropic reinstates OpenClaw, limits Agent credits

Anthropic has reversed a restriction it imposed in early April and is again allowing Claude subscribers to route their subscriptions through third-party agent tools such as OpenClaw, according to a May 13 VentureBeat report. The reinstatement does not restore the old economics: Anthropic now places that programmatic use inside a separate monthly “Agent SDK” credit bucket for paid subscribers, rather than letting it draw from the broader subscription pool. (venturebeat.com) That distinction matters because Anthropic had said the earlier ban was driven by capacity and cost. On April 3, Boris Cherny, Anthropic’s head of Claude Code, said the company’s subscriptions “weren’t built for the usage patterns of these third-party tools” and that Anthropic was prioritizing customers using its own products and API. VentureBeat reported that some users on $20 to $200 monthly Claude Pro and Max plans had been consuming far more model usage through external agent harnesses than those flat-rate subscriptions were designed to cover. (venturebeat.com) The practical change is that OpenClaw-style usage is no longer treated like ordinary interactive Claude use. VentureBeat reported that when a subscriber uses non-interactive or external agent workflows, that usage now counts against a fixed, non-rollover Agent SDK allowance. If a user assigns too much of their subscription value to that bucket and does not use it by month-end, it expires; if an agent is inefficient, it can burn through the budget faster. VentureBeat described the relevant subscription range as $20 for Claude Pro and $100 to $200 for Claude Max. (venturebeat.com) Anthropic’s own product and engineering materials help explain why the company is drawing a line between first-party and third-party agent use. The company says Claude Agent SDK gives developers access to the same core tools and context-management systems that power Claude Code, and Anthropic has separately published guidance on “effective harnesses for long-running agents,” including compaction and other techniques meant to manage context windows more efficiently. Anthropic also says Claude Code is built to operate across a project, edit files, run tests and use tools across a codebase. (anthropic.com) Anthropic has also been signaling that compute supply remains central to its product decisions. On May 6, the company said it had raised Claude usage limits and signed a compute partnership with SpaceX that would add more than 300 megawatts of capacity and improve service for Claude Pro and Max subscribers. That announcement did not mention OpenClaw specifically, but it shows Anthropic still framing subscription limits and product access around available infrastructure. (anthropic.com) For users, the story is less that OpenClaw is simply “back” than that Anthropic has changed the billing boundary. Third-party agents can again be connected to Claude subscriptions, based on VentureBeat’s report, but they now sit inside a metered monthly credit system aimed at programmatic use rather than the broader flat-rate experience subscribers had previously expected. Anthropic’s official newsroom listed no separate May 13 announcement about OpenClaw itself, so the clearest public description of the change remains VentureBeat’s reporting and Anthropic materials describing the Agent SDK and subscription products. (venturebeat.com)