Anthropic Resets All Pro/Max Usage Allowances Over Claude Code Bug
On June 1, 2026, Anthropic announced that it had fixed a bug in which some Claude Code sessions spawned excessive parallel subagents and consumed rate limits faster than expected, and that it had reset both the 5-hour limit and the weekly limit for all users on Pro/Max plans.
What Happened
The official developer account @ClaudeDevs explained that a bug had caused some Claude Code sessions to generate excessive parallel subagents, consuming usage faster than expected. After fixing the bug, the company reset both the 5-hour rolling window and the weekly rate limit for all affected users on Pro/Max plans.
According to a follow-up post in the thread, the issue was related to processing of Opus 4.8 requests, in which the model triggered more parallel tool calls than intended. It was also explicitly stated that this was unrelated to dynamic workflows.
Background and Significance
Claude Code is Anthropic's agentic coding tool, distinguished by its ability to launch subagents in parallel within a main session, each with independent context and tool sets. This parallelism is effective for task delegation and context management, but in heavy users' environments it can rapidly consume compute resources and tokens.
In the summer of 2025, Anthropic introduced a weekly limit in addition to the existing 5-hour limit in order to curb concentrated usage. According to TechCrunch's reporting, this primarily targeted top-tier heavy users on Pro/Max (less than 5% of all users). Limit resets have been carried out in the past for prompt caching bugs and excessive usage via third-party tools as well, and this response can be seen as part of ongoing stabilization measures. Compared to competitors such as Cursor and OpenAI Codex, the way Claude Code's subagent parallelism generates a unique consumption pattern stands out.
The plans available are Claude Pro ($20/month) and Claude Max ($100/month or $200/month), and usage can be checked via the /usage command, which shows a breakdown by skills/subagents/plugins/MCP server (official documentation).
Reactions
On X, immediately after the announcement, expressions of gratitude such as "Ty!" and "🔥🔥🔥" came one after another, and some users reported satisfaction, saying "Claude has good limits again" and "Even running multiple terminals and many subagents, I've still only used 14%."
On the other hand, there have been many complaints posted up to now—due to heavy use of subagents or use of third-party tools—such as "hit the limit in 5 minutes" and "ran out within an hour even on the Max plan." In the community, parallel subagents are recognized as the main driver of accelerated consumption, and saving tips such as using a single subagent, adjusting effort levels, and shifting time zones have been shared. Discussions over heavy-use cases and past bug fixes continue on HN and Reddit as well (example HN discussion).