The number hit the developer community like a verdict written in code: 67 percent. That is the measured drop in Claude Code's ability to handle complex engineering tasks after its latest update, documented in a viral GitHub issue that users quickly dubbed the "execution report"—written, with grim irony, by the model itself.
But the capability collapse was only the latest blow. Days earlier, Anthropic severed access to Claude Code subscriptions through third-party tools like OpenClaw. Developers can no longer use their paid subscriptions as a pass-through for external coding harnesses. Instead, they must pay-as-you-go or supply their own API keys. Anthropic offered one month of credits as compensation—a gesture many saw as insulting given the scale of what it had taken away.
The pattern behind these moves reveals something uncomfortable about Anthropic's trajectory. As the company's run-rate revenue hit $30 billion, up from $9 billion at the end of 2025, it has become increasingly aggressive about capturing compute spend directly. Third-party harnesses like OpenClaw consumed massive resources without delivering revenue to Anthropic's bottom line. So the company is steering users toward its own agentic tools—Dispatch, scheduled tasks, projects, and computer use—features that directly overlap with what OpenClaw offers.
"Agentic usage through third-party harnesses was eating massive compute," one developer wrote on X, summing up Anthropic's apparent logic. "They want that money."
The developer community is not taking this quietly. OpenClaw's founder, Peter—a developer known for pushing models to their limits—was already working on multi-model support. Within days of Anthropic's moves, he announced that OpenClaw would prioritize support for Alibaba's Qwen models, which had been dominating OpenRouter rankings. The Qwen pivot signals that developers are not waiting for Anthropic to rediscover its commitment to the ecosystem that helped build it.
For many developers, this feels like a betrayal. Claude Code was supposed to be a professional-grade tool, not a subscription product designed to funnel users into a corporate ecosystem. The 67 percent capability drop—confirmed by multiple benchmarks on complex refactoring, multi-file debugging, and architectural planning tasks—suggests that Anthropic either cannot maintain quality while scaling or has stopped trying.
The broader implication is troubling. When AI labs reach sufficient scale, they face a choice: remain open platforms that serve developers, or become walled gardens that extract maximum value from users. Anthropic appears to have made its decision. The question now is whether the developer community will follow OpenClaw to Qwen, or find yet another home.
OpenClaw's Qwen integration is already in beta. Developers who built their workflows around Claude Code are watching closely—and many are already moving.