Skip to content

Max 20 plan: rate limit 100% exhausted within ~70 minutes after reset (never happened before v2.1.89) #41788

@ArkNill

Description

@ArkNill

Environment

  • Plan: Max 20 ($200/month subscription)
  • Claude Code version: 2.1.89 (auto-updated today, April 1 2026)
  • Previous versions: 2.1.86 (Mar 28) → 2.1.87 (Mar 29) → 2.1.88 (Mar 31) → 2.1.89 (Apr 1)
  • Model: Claude Opus 4.6 (1M context)
  • OS: Ubuntu 24.04, Linux 6.14.0-1020-oem (HP ZBook Ultra G1a)
  • Node: v22.22.1
  • Settings: effortLevel: "high", 1 PostToolUse hook (ruff linter, lightweight)

Problem

Starting today (April 1, 2026), my 5-hour rate limit reaches 100% far too quickly. This happened twice in the same day.

This has never happened before. I have been a Max 20 subscriber for months and have never hit the limit this fast, even with identical (or heavier) usage patterns.

Timeline

Time Event
~08:30 KST Started working (remote session)
~11:30 KST Noticed rate limit already at 100% — roughly 3 hours of normal usage
12:00 KST Rate limit reset confirmed
12:00~ KST Resumed normal work immediately after reset
~13:10 KST 100% usage again — exhausted in ~70 minutes of light conversational work

What I was doing

  • Normal conversational coding tasks (not bulk/batch)
  • Working from home directory with project CLAUDE.md (~17KB) + MEMORY.md (~14KB) auto-loaded
  • No unusual tool-heavy operations, no massive file reads
  • Same workflow I've used daily for months without issues

What changed

  • v2.1.89 auto-updated today (Apr 1, 10:15 KST) — but the morning session (pre-update) also had the same issue
  • v2.1.88 installed Mar 31 — issue may have started then
  • No changes to my settings, hooks, or project files before this started

Investigation done

  1. Optimized CLAUDE.md + MEMORY.md from 30KB → 14KB (53% reduction) — did not resolve the issue
  2. Confirmed no heavy background processes or concurrent Claude Code sessions
  3. Checked GitHub issues — found related reports:

Expected behavior

With Max 20 plan and normal conversational usage, the 5-hour rate limit should last at least 3-4 hours of active use, as it has for the past several months.

Actual behavior

  • Morning: 100% in ~3 hours
  • Afternoon (after reset): 100% in ~70 minutes with even lighter usage

Reproducible across two consecutive reset cycles on the same day.

Possible causes

  • Server-side change in how cached/thinking tokens are counted against rate limits
  • Prompt cache invalidation causing full re-processing of system prompts each turn
  • Rate limit calculation regression in v2.1.88 or v2.1.89
  • Model change (Opus 4.6) affecting internal token accounting

Request

I am paying $200/month for the Max 20 plan, and the service has become practically unusable — 70 minutes of light usage per 5-hour window is not acceptable for a premium subscription. This is severely impacting my daily workflow and productivity. I would appreciate a prompt investigation and resolution of this issue. Thank you.

Related issues

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type
    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions