We need to talk about the Claude Code rate limits

Anthropic tightened Claude Code session limits during weekday peak hours (5:00 a.m. to 11:00 a.m. Pacific), pushing some subscribers through their 5-hour session caps faster than before. The change matters because it directly reduces how much compute paying users can reliably consume at the exact times when enterprise and heavy workloads tend to run—prompting widespread backlash after the adjustment was communicated late and only partially.

The new policy keeps weekly limits unchanged, but alters the distribution across the week: during weekdays in that morning Pacific window, users move through their 5-hour session limits faster than prior to the update. A key flashpoint is Anthropic’s claim that roughly 7% of users will hit session limits they wouldn’t have reached before, with the impact described as especially concerning for Pro tiers. Community reactions reflect the mismatch between expectations set by earlier “spring break” style promotions (including a temporary 2x usage outside the peak window) and the sudden tightening inside the peak window.

The frustration intensified because many users reported hitting limits immediately after the change went live—sometimes far earlier than their typical workflows would suggest. One account described using 100% of a 5-hour limit on a day when the same kind of work usually consumes only about 10%. Even where the exact magnitude is disputed, the optics were damaging: the timing of public confirmation was late, and the information appeared to be shared primarily via an individual employee’s social post rather than through official product surfaces like the dashboard, CLI, or official announcements.

Beyond the immediate rate-limit mechanics, the transcript frames the underlying driver as a broader “compute crisis.” GPUs are treated as a scarce, contested resource inside AI labs, with internal demand coming from three competing needs: research (training and model development), product (building and running Claude Code and related tooling), and users (subscription and API workloads). As customer growth accelerated—especially among high-spend enterprise accounts—product and user demand increasingly encroached on GPU capacity previously reserved for research.

The proposed economic logic is that Anthropic subsidized Claude Code compute heavily for a long time, including claims that some plans effectively enabled extremely high monthly compute consumption relative to price. When contention rose, the company faced a tradeoff: reduce overall access (which would likely harm revenue and retention) or reduce access only during the hours when GPUs are most contested. The chosen approach—lowering effective capacity during the 5:00 a.m. to 11:00 a.m. Pacific window—aims to protect enterprise and other high-value usage while preserving weekly totals.

Still, the transcript argues that Anthropic’s communication and internal prioritization culture amplify the backlash. The person discussing the change contends that the company’s research-first DNA and internal power dynamics make it harder to anticipate how users will experience GPU reallocations. The result is a policy that may be economically rational but feels abrupt, under-specified, and poorly communicated—especially when users can’t easily see the new constraints before they hit them.