Max 20x (00/mo) Opus 4 weekly token budget exhausted in 3 days of normal interactive use

Resolved 💬 2 comments Opened Mar 9, 2026 by triphase-physics Closed Mar 12, 2026

Description

I'm on the Max 20x plan ($200/month) and consistently hit ~70% of my weekly token allocation within 2-3 days of normal interactive use with Opus 4. This leaves me rationing tokens for the remaining 4+ days until the weekly reset.

My Use Case

  • Not running Claude Code in the background 24/7
  • Not using automated/agentic loops
  • Interactive sessions: reading source files, editing C code, debugging, compiling
  • Engineering work: writing C programs (~2000 line files), deploying builds
  • Physics research: reviewing technical documents, cross-referencing data
  • File management: reading/writing config files, updating documentation
  • Typical session: 2-4 hours of active work

The Problem

Opus 4 consumes tokens 3-5x faster than Sonnet per interaction. For users who need Opus-level reasoning (complex debugging, novel physics derivations, cross-domain pattern recognition), being told to "just use Sonnet" defeats the purpose of paying for the highest tier.

The Max 20x plan is marketed as the power-user tier with "20x the usage of Pro." But the weekly Opus 4 cap (reported as 24-40 hours) doesn't reflect actual interactive usage patterns where:

  1. Every message re-reads the full conversation history (token cost scales with conversation length)
  2. Reading source files (1000-2000 lines) costs ~30-60K input tokens per read
  3. Tool calls (file reads, searches, web fetches) each add to the token count
  4. The model's internal chain-of-thought reasoning (invisible to the user) consumes significant output tokens

What Would Help

  1. Increase the Opus 4 weekly allocation for Max 20x users — the current cap is too low for the advertised "power user" tier
  2. Better token visibility — show users their token consumption per message so they can budget. Currently there's no way to see how many tokens a single interaction cost.
  3. Separate input vs output token accounting — reading a large file (input) shouldn't count the same as model reasoning (output) against the weekly cap
  4. Carry-over or rollover — unused tokens from one week should roll into the next, rather than resetting to zero

Environment

  • Claude Code CLI on Windows 11
  • Model: claude-opus-4-6
  • Plan: Max 20x ($200/month)
  • Usage pattern: 2-4 hour interactive sessions, ~5 days/week

View original on GitHub ↗

This issue has 2 comments on GitHub. Read the full discussion on GitHub ↗