135,000 running instances. That's what it took for Anthropic to decide OpenClaw's users had outgrown their subscription plans.
On April 4, 2026, Anthropic blocked OpenClaw - an open-source AI agent framework - from running on Claude's flat-rate subscription plans. The problem: a single active OpenClaw instance could consume between $1,000 and $5,000 in equivalent API costs per day, while paying the same monthly rate as a casual user sending a few messages a week.
OpenClaw, originally called Clawdbot when Austrian developer Peter Steinberger launched it in November 2025, gives AI models persistent memory, tools, and integrations with messaging apps like WhatsApp and Telegram. It grew fast - 247,000 GitHub stars and 47,700 forks by early March 2026. At 135,000 active instances running continuously, it stopped being a side project and became a real infrastructure cost problem for Anthropic.
The Competitive Angle
The timing invites scrutiny. Steinberger joined OpenAI on February 14, 2026, with the company backing OpenClaw's transition to an open-source foundation. Anthropic announced and enforced the subscription restrictions within weeks - which prompted criticism that the policy targeted a popular framework now affiliated with a direct competitor.
Anthropic's stated reason is straightforward: "Anthropic's subscriptions weren't built for the usage patterns of these third-party tools." That's defensible. Flat-rate pricing and autonomous agents running around the clock are structurally incompatible. But the timing does nothing to quiet skeptics.
For affected users, Anthropic is offering one-time credits equal to their monthly plan cost, redeemable until April 17, plus 30% discounts on prepurchased usage bundles. Under the new pay-as-you-go model, per-task costs run $0.50 to $2.00 - between 10x and 50x more expensive per month for heavy users depending on workload.
The people hit hardest are hobbyist developers and solo practitioners, not enterprise teams with budget flexibility to absorb the change.