claude-code - 💡(How to fix) Fix [BUG] Investigate usage limits on Max 20x subscriptions [1 comments, 2 participants]

Root Cause

Bug Description You most definitely have a bug in input/output or cache because I hit my limits just from 2-3 research tasks, with crashing agents and junk 4.1m token output that is unusable. Also I found out I used ~90$ of extra usage only after I restarted the session. Something is off please investigate and if more information

Code Example

Current session
  Resets 5pm (Europe/hidden)█████████████████████ 100% used

  Current week (all models)
  Resets Apr 24 at 8am (Europe/hidden)            31% used


  Current week (Sonnet only)
  Resets Apr 24 at 8am (Europe/hidden)

  Approximate, based on local sessions on this machine — does not include other devices or claude.ai


  Last 24h · these are independent characteristics of your usage, not a breakdown
  95% of your usage came from subagent-heavy sessions
   Each subagent runs its own requests. Be deliberate about spawning them — and
   consider configuring a cheaper model for simpler subagents.
  35% of your usage was while 4+ sessions ran in parallel
   All sessions share one limit. If you don't need them all at once, queueing uses
    it more evenly.
  23% of your usage was at >150k context
   Longer sessions are more expensive even when cached. /compact mid-task, /clear
   when switching to new tasks.

  d to day · w to week

  Extra usage
  $83.76 / $170.00 spent · Resets May 1

---

Current session
  Resets 5pm (Europe/hidden)█████████████████████ 100% used

  Current week (all models)
  Resets Apr 24 at 8am (Europe/hidden)            31% used

  Current week (Sonnet only)
  Resets Apr 24 at 8am (Europe/hidden)            15% used

  Approximate, based on local sessions on this machine — does not include other devices or claude.ai

  Last 7d · these are independent characteristics of your usage, not a breakdown

  91% of your usage came from subagent-heavy sessions
   Each subagent runs its own requests. Be deliberate about spawning them — and
   consider configuring a cheaper model for simpler subagents.
  48% of your usage was while 4+ sessions ran in parallel
   All sessions share one limit. If you don't need them all at once, queueing uses
    it more evenly.
  34% of your usage was at >150k context
   Longer sessions are more expensive even when cached. /compact mid-task, /clear
  25% of your usage came from sessions active for 8+ hours
   These are often background/loop sessions. Continuous usage can add up quickly
   so make sure it is intentional.
  d to day · w to week

  Extra usage
  $83.76 / $170.00 spent · Resets May 1 (Europe/hidden)

The below session details involved research only, with failed agents that crashed and apprently still generated output. The orchestrator was the only one on opus and didn’t even go above 170k token usage. My sub-agents were all setup to use sonnet and haiku, why were they overwritten to use opus again? A lot of weird things happening and also the orchestrator was setup on medium effort.

  Current session
  Resets 5pm (Europe/hidden)█████████████████████ 100% used

  Current week (all models)
  Resets Apr 24 at 8am (Europe/hidden)            31% used


  Current week (Sonnet only)
  Resets Apr 24 at 8am (Europe/hidden)

  Approximate, based on local sessions on this machine — does not include other devices or claude.ai


  Last 24h · these are independent characteristics of your usage, not a breakdown
  95% of your usage came from subagent-heavy sessions
   Each subagent runs its own requests. Be deliberate about spawning them — and
   consider configuring a cheaper model for simpler subagents.
  35% of your usage was while 4+ sessions ran in parallel
   All sessions share one limit. If you don't need them all at once, queueing uses
    it more evenly.
  23% of your usage was at >150k context
   Longer sessions are more expensive even when cached. /compact mid-task, /clear
   when switching to new tasks.

  d to day · w to week

  Extra usage
  $83.76 / $170.00 spent · Resets May 1

Most of the sub-agents crashed, which means when your agents crash they still generate output. Output which can be lost or of poor quality and unsable when recovered. This started after the update once 4.7 was released

Now look at the usage per week where only 48% was used (including the session today) with 4+ sessions prior to release

  Current session
  Resets 5pm (Europe/hidden)█████████████████████ 100% used

  Current week (all models)
  Resets Apr 24 at 8am (Europe/hidden)            31% used

  Current week (Sonnet only)
  Resets Apr 24 at 8am (Europe/hidden)            15% used

  Approximate, based on local sessions on this machine — does not include other devices or claude.ai

  Last 7d · these are independent characteristics of your usage, not a breakdown

  91% of your usage came from subagent-heavy sessions
   Each subagent runs its own requests. Be deliberate about spawning them — and
   consider configuring a cheaper model for simpler subagents.
  48% of your usage was while 4+ sessions ran in parallel
   All sessions share one limit. If you don't need them all at once, queueing uses
    it more evenly.
  34% of your usage was at >150k context
   Longer sessions are more expensive even when cached. /compact mid-task, /clear
  25% of your usage came from sessions active for 8+ hours
   These are often background/loop sessions. Continuous usage can add up quickly
   so make sure it is intentional.
  d to day · w to week

  Extra usage
  $83.76 / $170.00 spent · Resets May 1 (Europe/hidden)

These issues and instabilities force me to switch to other subscriptions and providers for a while.

Environment Info

Platform: darwin
Terminal: ghostty
Version: 2.1.114
Feedback ID: b92ffa30-3a21-4ce5-8bc5-f51a4cd6802f

<!-- Uploading "Screenshot 2026-04-19 at 15.51.23.png"... —>

extent analysis

TL;DR

Consider reconfiguring subagents to use cheaper models and limiting parallel sessions to reduce usage and prevent crashes.

Guidance

Review subagent configurations to ensure they are using the intended models (e.g., Sonnet and Haiku) instead of Opus, which may be causing increased usage.
Limit the number of parallel sessions, as 35% of usage came from 4+ sessions running simultaneously, to prevent exceeding the shared limit.
Monitor and adjust context sizes, as 23% of usage was at >150k context, which can lead to increased costs even with caching.
Consider implementing queueing for sessions to use resources more evenly and prevent crashes.

Example

No code snippet is provided, as the issue is more related to configuration and usage patterns.

Notes

The issues seem to be related to the update to version 4.7, and the provided data suggests that subagent-heavy sessions and parallel sessions are contributing to the increased usage. However, without more information about the specific configurations and usage patterns, it's difficult to provide a more detailed solution.

Recommendation

Apply workaround: Reconfigure subagents and limit parallel sessions to reduce usage and prevent crashes, as this seems to be the most likely cause of the issues.

Data

Security

Network

Code

UI/UX

Text

System

Multimedia

Protocol

API

Engineering

claude-code - 💡(How to fix) Fix [BUG] Investigate usage limits on Max 20x subscriptions [1 comments, 2 participants]

Recommended Tools

GitHub issue graph ai analysis

Root Cause

Code Example

extent analysis

TL;DR

Guidance

Example

Notes

Recommendation

Still need to ship something?

TRENDING

claude-code - 💡(How to fix) Fix [BUG] Investigate usage limits on Max 20x subscriptions [1 comments, 2 participants]

Recommended Tools

GitHub issue graph ai analysis

Root Cause

Code Example

extent analysis

TL;DR

Guidance

Example

Notes

Recommendation

Still need to ship something?

RELATED_DISCOVERY

TRENDING