Claude Code Quota Crisis: Bugs and Policy Changes Exhaust User Limits

Anthropic is investigating why Claude Code users are hitting their usage limits much faster than expected, with some reporting their quotas are exhausted in just one hour. The issue appears to be a combination of recent policy changes, the end of a promotion, and potential software bugs that break prompt caching and inflate costs. This disruption highlights the growing conflict between AI vendors' marketing of automated workflows and the restrictive nature of their quota systems.

Key Points

• Anthropic is investigating reports that Claude Code users are hitting usage limits in as little as one hour, compared to previous eight-hour sessions.
• Potential causes include the end of a usage promotion, new peak-hour restrictions, and software bugs that break prompt caching and inflate costs by up to 20x.
• The default five-minute prompt cache lifetime forces higher costs on developers who take even short breaks during their coding sessions.
• Anthropic's lack of transparent, hard usage limits for Pro and Team plans makes it difficult for developers to plan or budget their work effectively.
• Automated workflows are particularly at risk, as rate-limit errors can trigger retry loops that drain an entire daily budget in minutes.

Sentiment

The community overwhelmingly sides with the article's framing. There is near-universal frustration with Anthropic's opaque quota system and the sudden degradation of service. While a few voices defend the economic realities of AI pricing or note they haven't personally been affected, the dominant sentiment is one of anger, distrust, and active exploration of alternatives. Many commenters go further than the article, accusing Anthropic of deliberately manipulating quotas rather than simply having bugs.

In Agreement

• Multiple users confirm hitting limits after minimal usage — one Max plan user hit 12% quota by typing 'test one two three', another reached 20% by typing 'continue'
• Technical analysis confirmed a cache invalidation bug where a sentinel string embedded in the request body can be corrupted by billing-related conversation content, forcing expensive full-context reprocessing
• The expiration of the 'double usage' promotion combined with reduced peak-hour quotas creates a compounding perception of degraded service
• Anthropic's customer support via the AI chatbot 'Fin' is widely regarded as useless, with no viable path to reach a human for billing disputes
• The lack of transparency around exact quota limits and token accounting makes it impossible for developers to plan their work sessions
• Users drawing on API keys report seeing unexpectedly high costs that corroborate the subscription quota complaints

Opposed

• Some users report no change in their usage patterns, suggesting the bug may affect only certain accounts or configurations — indicating possible A/B testing or inconsistent bug triggers
• Several commenters argue users have unrealistic expectations about what unlimited AI compute should cost at $20/month
• The identified cache bug only affects conversations containing specific sentinel strings, which shouldn't be common for most users
• Some users find Claude Code still delivers enough value on the Max plan to justify the cost despite the quota issues
• Technology will continue improving — hardware efficiency, model optimization, and better agent design will reduce costs over time