OpenClaw 通过分析 Anthropic 切断 OAuth 事件，展示了无缝切换多模型能力及接口层的主导地位。

OpenClaw's tweet is the most interesting thing happening in AI right now and almost nobody is reading it correctly.

350,000 GitHub stars. 3.2 million active users. The most-starred software project in GitHub history, passing React in 60 days. And the project just got cut off by its primary model provider two days ago.

Here's what actually happened. Anthropic killed OAuth access for Claude Pro and Max subscribers using third-party tools on April 4. A single OpenClaw instance running 24/7 burns $1,000 to $5,000 in equivalent API costs per day. Users were paying $20/month. The unit economics were so underwater that Anthropic's Head of Claude Code personally submitted pull requests to OpenClaw's codebase trying to fix the caching problem before they pulled the plug.

That last part is the detail that tells you everything. Anthropic tried to engineer its way out of cutting off its own ecosystem. The math won anyway.

Now look at OpenClaw's response. "Anthropic cut us off. GPT-5.4 got better. We moved on." Three sentences. No anger. No campaign. They shipped video generation, music generation, and 12 new languages in the same release. The project had already added OpenAI Codex, Alibaba Cloud, MiniMax, xAI, and Amazon Bedrock as providers.

This is the real lesson buried in this release. OpenClaw doesn't need Claude. Claude needed OpenClaw's 3.2 million power users choosing it as their default model. Steinberger built the project on Claude, got trademark complaints, got his OAuth killed, and is now an OpenAI employee shipping GPT-5.4 compatibility into the codebase.

The AI platform war just revealed its actual structure. Models are commoditizing. The interface layer is where lock-in lives. And the project with 350,000 stars just proved that the switching cost between frontier models is approximately one config file change and a Friday afternoon.