Anthropic prevents OpenClaw from subscribing to Claude in a cost-reduction effort | TNW

      In summary: Anthropic has restricted Claude Pro and Max subscribers from using flat-rate plans with third-party AI agent frameworks, starting with OpenClaw, effective April 4, 2026. This change shifts the financial burden of running autonomous agents to users through a pay-as-you-go billing model. The creator of OpenClaw, who joined OpenAI in February, criticized the decision as a betrayal of open-source developers, leading to cost increases for thousands of users, with some facing up to a 50-fold rise in their monthly expenses.

      Anthropic has eliminated a previously unpublicized subsidy that made its Claude models the preferred choice for the open-source AI agent community. As of April 4, 2026, users of Claude's Pro and Max subscription tiers can no longer utilize their plan's usage limits with third-party frameworks like OpenClaw. To continue using these tools with Claude, they will now incur additional costs under a new "extra usage" billing format. Anthropic has indicated that it will extend this restriction to all third-party integrations in the future.

      This announcement shocked many developers who had based their AI setups on the assumption that a flat monthly fee would suffice. For a lot of users, that’s no longer the case.

      The rationale behind this shift is clear, although the timing was unexpected. Claude's subscription plans were created for conversational use—when a user initiates a chat, asks a question, and receives an answer. However, agentic frameworks operate quite differently. A single OpenClaw instance running continuously for one day, performing tasks like web browsing, managing schedules, and responding to messages, can incur API costs ranging from $1,000 to $5,000, depending on the workload. Under a $200 monthly Max subscription, this represented an unsustainable cost shift from users to Anthropic.

      “Anthropic's subscriptions weren’t designed for the usage patterns of these third-party tools,” stated Boris Cherny, Head of Claude Code at Anthropic. “We manage capacity as a resource thoughtfully and prioritize customers using our products and API.”

      The scale of the issue was considerable, with over 135,000 OpenClaw instances estimated to be in operation at the time of the announcement. Industry analysts noted a significant price disparity, with heavy agentic users paying over five times less under flat subscriptions compared to equivalent API rates. Effectively, Anthropic's subscription model had been quietly subsidizing a class of usage it had not appropriately priced.

      OpenClaw is an open-source AI agent framework developed by Austrian programmer Peter Steinberger. Initially released in November 2025 as Clawdbot, it started as a side project where Steinberger aimed to explore the potential of giving a large language model persistent memory, tool access, and communication capabilities through messaging apps like WhatsApp and Telegram. This project garnered immense interest, as many users desired such functionality.

      The project underwent two name changes within three days in late January 2026: first to Moltbot, after Anthropic raised trademark issues about the similarity to "Claude," and then to OpenClaw three days later. By March 2, 2026, the repository had garnered 247,000 GitHub stars and 47,700 forks, becoming what many sources termed the fastest-growing GitHub project in history, hitting 100,000 stars in less than 48 hours at its peak. The framework supports more than 50 integrations and operates across Claude, GPT-4o, Gemini, and DeepSeek. Tencent even built an enterprise platform on it, illustrating that OpenClaw's impact reached well beyond individual hobbyists.

      The timing of the restriction is more significant given events in February. On February 14, 2026, Steinberger announced his departure from his project to join OpenAI, with Sam Altman stating that Steinberger would “drive the next generation of personal agents” at OpenAI and that OpenClaw would transition to an open-source foundation with continued support from the company. Steinberger expressed that collaborating with OpenAI was the quickest way to make this accessible to everyone.

      Anthropic's restrictions emerged just weeks after this announcement, a timeline that hasn't gone unnoticed. Steinberger and fellow investor Dave Morin sought to negotiate a more favorable outcome, reaching out to Anthropic, but they claimed they only succeeded in delaying enforcement by a week.

      "Initially, they replicate some popular features in their closed framework, and then they exclude open-source," Steinberger responded to the ban.

      Whether this timing is indicative of a competitive strategy or mere coincidence, the outcome remains the same: the most prominent open-source agent framework, now loosely connected with OpenAI, has effectively been priced out of Claude's subscription offerings.

      For developers used to unlimited agentic operations under a flat plan, the new billing structure presents a substantial disruption. Under the pay-as-you-go extra usage model, costs per interaction are expected to be between $0.50 and $2.00 per task, making frequent