April, When All AI Subscription Plans Collapsed

In April, especially within 3 weeks, 5 announcements came in succession from Anthropic, OpenAI, and GitHub. Subscription plans designed around chat could no longer handle long-running AI agents and automated workflows.

• On April 4, with less than 24 hours' notice, Anthropic blocked Claude Pro/Max from using third-party agent execution tools like OpenClaw. Claude Code and Cowork maximize prompt cache hits, processing cached input tokens at 10% of the standard rate, but third parties couldn't match that pattern, incurring costs of around $1,000–$5,000 per day.
• On April 9, OpenAI launched a $100/month Pro 5x tier aimed at heavy Codex users.
• On April 13, GitHub discontinued new Copilot Pro trials, citing abuse.
• On April 16, Anthropic released Opus 4.7 with a new tokenizer. API pricing stayed the same as Opus 4.6, but the same input was now counted as up to 35% more tokens.
• Between April 20–23, GitHub blocked new sign-ups for Copilot Pro, Pro+, and Student, leaving only free Copilot as the new individual sign-up path. Opus was removed from Pro, Opus 4.5/4.6 were removed from Pro+, and Opus 4.7 was given a 7.5x request multiplier. Anthropic briefly removed Claude Code from Pro before reversing the decision, and OpenAI raised GPT-5.5 API pricing from $2.50/$15 per million tokens to $5/$30.

As Anthropic's Head of Growth Amol Avasare put it, Max was originally designed for chat-centric use and didn't account for Claude Code, Cowork, or hours-long async agents. In a structure where all users are tied to the same OAuth token and the same inference path, it's hard to finely distinguish between chat and autonomous agents. Integration tools that hardcode a fixed multiplier inevitably overcharge users if they miss the new tokenizer reality. In the end, this episode isn't about model competition—it's a case where billing engineering failed to keep up with product evolution.