Anthropic throttles Claude rate limits, devs call foul

In a significant policy shift, Anthropic has announced that it will impose weekly usage limits for subscribers of its AI model, Claude, starting August 28. The company claims this decision comes in response to certain users running Claude continuously, particularly its Claude Code feature, which has seen a surge in demand. The new weekly limits will be introduced alongside the existing 5-hour daily caps. Anthropic reassured that only 5% of its user base would be impacted by this throttling. Nonetheless, the announcement has drawn considerable criticism from developers who feel the move punishes the majority for the misuse by a few. "Claude Code has experienced unprecedented demand since launch. While most users operate within standard usage patterns, we have also encountered issues such as account sharing and unauthorized access reselling, which detrimentally affects overall performance," Anthropic stated in a communication to Venture Beat. They noted that some users have engaged in intensive usage, such as running Claude 24/7, which has strained system capacity. To maintain service reliability, Anthropic indicated that it is exploring additional options for prolonged usage in the future, but for now, these weekly limits aim to create a more stable environment for all users. Although specific rate limits were not disclosed, the company mentioned that users of Claude Max 20x could anticipate between 240 to 480 hours of Sonnet 4 and 24 to 40 hours of Opus 4 within their weekly allowance. Heavy users of the Opus model or those operating multiple instances of Claude Code may find themselves reaching these thresholds sooner than expected. For those who do exceed their limits, the option to purchase additional usage at standard API rates will be available to ensure uninterrupted work. Some enterprises might already have existing agreements concerning these rate limits, yet others may find themselves needing to upgrade their subscriptions to meet project demands. This policy change arrives amid reports of reliability issues with Claude, which Anthropic acknowledges and is actively working to resolve. As Anthropic continues to make strides in the developer community, it is also advocating for the broader acceptance of AI coding tools. Earlier this year, the company transformed Claude into a no-code platform and launched a financial services-specific version for enterprise users. While some competitors, like Google, have begun to remove limitations on certain models, Anthropic and OpenAI continue to offer tiered rate limits, allowing users to pay more for increased computational resources. Despite the rationale behind these rate limits, many paying users are expressing dissatisfaction, feeling that they are being penalized for the actions of a few who exploit the system. The debate continues as users weigh the necessity of these limits against the potential hindrance to their projects, especially for those experimenting with long-running applications or engaging in extensive coding tasks.