Anthropic has reset the five-hour and weekly limits for all Claude Code Pro and Max subscribers after a software bug caused some sessions to burn through quotas much faster than expected. The company says the glitch has been fixed, and the reset is meant to give affected users back the capacity they lost when the system misbehaved.
The issue affected Claude Code sessions, not pricing or policy, and it came from a bug that triggered too many parallel subagents and tool calls. For paid users, that meant quotas could disappear after just a few requests instead of lasting through a normal coding session.
What went wrong in Claude Code
According to Anthropic, the fault sat in the request-handling logic for Claude Opus 4.8. Instead of keeping tool use within the intended bounds, the system triggered more simultaneous calls than planned. That made some sessions behave like they had caffeine and a grudge, racing through allotted usage on even relatively simple prompts.
Users noticed quickly and complained across social platforms, with some saying daily allowances disappeared after only a few requests. The blame briefly shifted in community chatter to recent changes in Dynamic Workflows, but Anthropic pushed back on that theory and said the limits issue was unrelated to that feature.
Claude Code limits reset for Pro and Max subscribers
For Anthropic, the cleanest fix was also the most visible one: restore the limits and move on. That is a sensible move in a market where developers can switch between coding assistants with very little friction, and where trust is measured in usage caps as much as model quality.
- Who was affected: Claude Code Pro and Max subscribers
- What was reset: five-hour and weekly limits
- What caused it: a bug that launched too many parallel subagents and tool calls
- What Anthropic says now: the issue has been fixed
The bigger question is how often AI coding tools will trip over their own orchestration layers as they get more aggressive about parallelism and automation. Anthropic is hardly alone here: every major assistant vendor is trying to squeeze more useful work out of each session, and every extra layer of automation adds another way to accidentally torch a user’s quota.
If the fix holds, the reset should soften the immediate fallout. If not, developers will do what they always do: compare notes, post screenshots, and start wondering which assistant is safest to trust with the next long coding session.