Anthropic Tests Dropping Claude Code from Pro

🔑 Enhanced Key Takeaways

• •The testing phase involved a temporary 'usage-based' throttle that restricted Claude Code access for Pro subscribers who exceeded specific daily token thresholds, rather than a blanket removal.
• •Anthropic's infrastructure strain is attributed to the high computational cost of Claude Code's multi-step reasoning capabilities, which require significantly more inference cycles than standard chat interactions.
• •Internal reports suggest Anthropic is evaluating a tiered 'Power User' add-on model to monetize high-volume coding agents separately from the standard Pro subscription.

🛠️ Technical Deep Dive

• •Claude Code utilizes a specialized agentic loop that performs iterative file system analysis, terminal command execution, and multi-file refactoring.
• •The system relies on a high-context window (up to 200k tokens) to maintain repository awareness, which significantly increases the KV cache memory footprint per request.
• •Anthropic implemented a 'cost-per-turn' monitoring system to track the compute-intensive nature of recursive tool calls, which are the primary drivers of the current service degradation.

🔮 Future ImplicationsAI analysis grounded in cited sources

⏳ Timeline