Claude just changed overnight

TL;DR

Anthropic blocked third-party tools like OpenClaw from using subsidized subscription tokens, triggering accusations of copying open-source features before locking out the community.

Key Points

• 1.Anthropic cut off third-party harnesses from subsidized subscription tokens on April 4th. Tools like OpenClaw could let users burn $200+/month in API costs while paying only $20–$200 for a Claude subscription, breaking Anthropic's business model.
• 2.The official justification was prompt cache inefficiency. Claude Code and Claude Codework were engineered to reuse cached context, dramatically reducing compute costs, while OpenClaw ignored or underutilized these optimizations, making it far more expensive to serve.
• 3.Anthropic introduced an 'Extra Use' API token option at up to 30% off, but even mentioning OpenClaw in a system prompt blocked Claude Code from running. This effectively forced third-party users onto metered per-token billing rather than flat-rate subscriptions.
• 4.Critics accuse Anthropic of a 'copy then close' strategy. OpenClaw creator Peter Steinberger says Anthropic heavily borrowed open-source community innovations — like dreaming/memory consolidation — into Claude Code, then locked out the open-source ecosystem that created them.
• 5.Peter Steinberger was acqui-hired by OpenAI, and OpenAI's Codex explicitly welcomed third-party harness use. The creator has hinted he may bring OpenClaw-style agent capabilities to GPT models from inside OpenAI.
• 6.The backlash is hitting Anthropic's most loyal evangelists hardest. Power users who drove word-of-mouth for Claude are now publicly criticizing the company, compounding recent bad optics including the Claude Mythos leak, source code leak, and a DMCA takedown of 8,000+ GitHub repos.