Max plan: usage exhausted abnormally fast, twice in one day (March 31)

Status Open
Maintainer reply None cached
Activity 10 comments · opened Mar 31, 2026

Description

I'm on Claude Code Max subscription. Today (March 31, 2026) my usage was exhausted abnormally fast — twice in one day. I barely did any work and the quota was already gone both times.

This is NOT normal behavior. Previously I could work extensively without hitting limits. Today the usage runs out almost immediately.

Environment

  • Plan: Claude Code Max
  • Model: Claude Opus 4.6 (1M context)
  • Platform: macOS (Darwin 24.6.0)
  • Claude Code: VS Code extension

Expected behavior

Usage should last for meaningful work sessions, as it did before.

Actual behavior

Usage exhausted within minutes of starting work. Happened twice today (March 31, 2026).
This started recently (late March 2026).

Request

Please review my account usage logs — the consumption pattern today is clearly abnormal compared to my previous usage history.

Related issues

#38335, #37394, #41424, #41212

View original on GitHub ↗

10 Comments

github-actions[bot] · 5 months ago

Found 3 possible duplicate issues:

  1. https://github.com/anthropics/claude-code/issues/41424
  2. https://github.com/anthropics/claude-code/issues/37394
  3. https://github.com/anthropics/claude-code/issues/38335

This issue will be automatically closed as a duplicate in 3 days.

  • If your issue is a duplicate, please close it and 👍 the existing issue instead
  • To prevent auto-closure, add a comment or 👎 this comment

🤖 Generated with Claude Code

alexsapunov · 5 months ago

Usage data — clear evidence of abnormal consumption

Here is my actual usage from the Anthropic dashboard:

| Date | Cost | Messages | Cache Creation | Cache Read |
|------|------|----------|---------------|------------|
| 31.03 (today) | $54.60 | 212 | 5,873,410 | 106,089,848 |
| 30.03 | $143.41 | 1,016 | 7,260,860 | 379,166,552 |
| 29.03 | $135.28 | 1,248 | 2,208,119 | 417,572,985 |
| 28.03 | $80.19 | 1,223 | 1,898,943 | 231,710,747 |
| 27.03 | $130.15 | 935 | 7,462,367 | 348,661,083 |
| 26.03 | $110.83 | 1,744 | 2,710,846 | 332,315,336 |

Today I hit the limit after only 212 messages — normally I send 900-1,700 messages per day.

The Cache Creation tokens today (5.8M) are disproportionately high relative to the number of messages (212), while Cache Read is very low (106M vs typical 200-400M). This strongly suggests the prompt cache is not being reused properly, causing full reprocessing on each turn.

This is 5-8x less capacity than my normal daily usage. Something is clearly broken with token accounting or cache handling today.

Please review my account logs.

flyte · 5 months ago

@alexsapunov how did you gather this usage data? The usual analytics stuff doesn't seem to track my Max subscription's usage.

alexsapunov · 5 months ago

@flyte The usage data comes from the Anthropic Console (not claude.ai).

Go to: https://console.anthropic.comUsage tab

There you can see daily breakdown with:

  • Input Tokens
  • Output Tokens
  • Cache Creation tokens
  • Cache Read tokens
  • Messages count
  • Cost

This is where I got the table from. It shows the internal cost Anthropic attributes to your Max subscription usage, even though you're not paying per-token.

The key metric to watch is Cache Creation tokens ÷ Messages — this gives you tokens-per-message. On a normal day mine is 1.6k-1.8k, but on the bug day (March 31) it spiked to 27.7k per message — a 17x increase.

flyte · 5 months ago

When I log in there I get a different organisation, so I'm not able to see the stats from my Max sub org. Not sure how you've been able to link them up!

alexsapunov · 5 months ago

@flyte When you log into console.anthropic.com, check the organization selector in the top-left corner. You should see something like "your-email's Organization" — that's the one linked to your Max subscription.

If you see a different org (like a work/team org), switch to your personal one. The Max subscription usage should appear under your personal organization.

Also make sure you're logging in with the same email that your Max subscription is billed to.

ArkNill · 5 months ago

Same experience here on Max 20 ($200/mo) with Opus — rate limit 100% exhausted in ~70 minutes.

I set up a monitoring proxy via the ANTHROPIC_BASE_URL env var and traced it to two client-side cache bugs:

  • Sentinel bug (#40524)
  • Resume bug (#34629) — deferred_tools_delta introduced in v2.1.69

The cache read ratio dropped to 4.3% (effectively ~20x token inflation). After workarounds it recovered to 89–99%.

Full analysis + workarounds: https://github.com/ArkNill/claude-code-cache-analysis

Important caveat: All my testing was done in the CLI environment (claude in terminal, Linux). Since you're on VSCode (macOS), the cache behavior might differ. That said, @Codename-404 reported in #38335 that downgrading the VSCode extension to v2.1.68 also significantly reduced consumption — so the same root cause likely applies.

You can try: VSCode extension → gear icon → "Download specific version VSIX" → select 2.1.68, and monitor if it helps.

ArkNill · 5 months ago

Follow-up — precautions (April 2, 2026)

"Exhausted twice in one day" = consistent with the cache bug compounded by session behaviors.

Precautions that help:

  • Avoid: --resume (replays full history), /dream & /insights (background drain), v2.1.89
  • Careful: sub-agents (Haiku = 0% cache read), multiple terminals
  • Use: v2.1.81 fixed + fresh sessions only + proxy monitoring

Ref: https://github.com/ArkNill/claude-code-cache-analysis

ArkNill · 5 months ago

Update (April 2): v2.1.90 has significantly improved cache efficiency — benchmark shows 95-99% cache read in stable sessions (both npm and standalone installations).

If you're still affected:

  1. Update: claude update (or npm install -g @anthropic-ai/claude-code)
  2. Pin the version: add "DISABLE_AUTOUPDATER": "1" to ~/.claude/settings.json env section
  3. Avoid --resume (still broken)

Note: server-side quota issues (org-level pool sharing, accounting mismatches) remain unresolved — the above fixes the client-side cache drain only.

Benchmark data: https://github.com/ArkNill/claude-code-cache-analysis

ArkNill · 5 months ago

April 3 update: v2.1.91 fixes the cache regression that caused the worst drain. If you are still hitting limits after updating, there are additional unfixed mechanisms: a 200K tool result budget cap, a client-side false rate limiter, and silent context stripping — all confirmed via proxy testing. Anthropic acknowledged peak-hour tightening on X (Lydia Hallie) but stated "none were over-charging you." Measured data and analysis: claude-code-cache-analysis