Anthropic tackles a frustrating issue with Claude Code, its popular AI coding assistant. Users hit token usage limits much faster than anticipated. The company investigates and calls this fix their top priority.
Developers purchase tokens for AI tasks, but consumption rates often surprise them. Reddit users report paid accounts ($100/month) burn through limits quicker than free ones. One session trapped in a code loop drained a daily budget in minutes.
Simple replies spike usage dramatically—one sentence jumped a user from 59% to 100%. Peak-hour throttling, added last week, accelerates token burn during high demand. These glitches disrupt developers’ daily workflows.
Claude Pro costs $20 monthly, with higher tiers at $100-$200. Businesses access custom pricing. Anthropic recently leaked 500,000 lines of internal Claude Code via human error on GitHub—no customer data exposed.
The firm battles the US government legally over Department of Defense tool usage. Independent developers previously reverse-engineered Claude Code; an earlier leak hit in February 2025.
