[BUG] Abnormal rate limit consumption — 1-2 prompts exhausting 5-hour window

I encountered the same problem; I used up my 5-hour limit on the Max 5X Plan after only two rounds of conversations. It's unbelievable.

Same issue here. Does it have anything to do with CLI version or model related?

1-2 prompts exhausting the limit sounds like the cache_creation bug others have reported in #38350. The token accounting seems broken.

While waiting for a fix, multi-model routing helped me — send the simple stuff through GPT-4o/Gemini so your Anthropic quota lasts the full session. I went from hitting limits daily to not hitting them at all.

Setup and details here if useful: https://dev.to/sophiaashi/claude-code-rate-limit-on-max-plan-the-workaround-developers-are-actually-using-in-2026-16g2

We also have a Discord for people dealing with this: https://discord.gg/tvAtTj2zHv

If you're on CC >=2.1.80, the stdin JSON already has rate_limits.five_hour.used_percentage and resets_at -- you can parse these in a statusline script to see the numbers in real time. I do this in claude-lens and the pace does seem off recently; a few prompts can spike the percentage way more than expected. At minimum, having the raw data visible helps you screenshot evidence when filing reports like this.

I encounter the same issue, 2 prompts and my 5 hour limit is gone. Im using claude code with opus on medium thinking

same issue with my pro sub, not just burning through current session usage but also extra usage credit in 1-2 promt. anyone found a quick solution instead of waiting fro 4-5hrs?

Just piling on, sonnet used an entire 5 hour quota (first prompt of the morning, so full period reset) asking it to fix a clippy warning. Didn't even finish the task and ran out of tokens in less than 5 minutes.

Codex cleaned it up in a a couple minutes with no ridiculous token consumption.

Same thing for me. I asked one question in Claude Code, went from 0 to 27%. That's not typical.

I found the reason, take a look at this post about the problem, it basically means that during peak hours anthropic has tweaked the usage to burn fast in order to discourage usage during peak hours. Honestly I think its aa terrible decision.

https://www.theregister.com/2026/03/26/anthropic_tweaks_usage_limits/

5x Max, non-peak, v2.1.87, Opus 4.6 (1M context effort high), macOS

Same issue. Two data points from a single session (2026-03-30 05:00 GMT-4 , non-peak hours):

Data point 1: 2 lightweight questions (settings review with ~20 file reads/bash commands + web search via 2
Sonnet sub-agents totaling ~74K tokens) → ~10% of 5h window consumed, context utilization only 6%.

Data point 2: 1 trivial follow-up (3 tool calls: 1 failed CLI command, 1 ls, 1 file write, ~100 output tokens)
→ 1% of 5h window consumed.

1% of a 5x Max non-peak 5h window for 3 tool calls is not explainable by the off-peak promotion ending or any
reasonable token accounting.

I just opened my claude in new 5hour window and it already says 10% used, what???

Yeh, im hitting abnormal usage too...
Normally 1x 5 hour limit is 8% for my total week.. During promo it was 5-6%...
Today? Its 14 fking percent... Thats HALF my usage per week...
What the actual fk is going on here???????
Max20 here...
Now im also getting this...
Ive always, for over a year worked in 10+ terminals without a single issue...
Its also not consistent, sometimes its on 7 terminals sometimes on 10... the only thing that is consistent is that it kills ALL MY terminals AND ALL my work.... blanket on every single client...
"API Error: Rate limit reached".... Cant even find info on it...

The same bug. It started around March 26–28

i've consumed 69% of my weekly in 3x 5 hour sessions, its been 1.5 days since reset... at this rate? its all gone tomorrow with sunday to wait... At worst? normally get to wait 12-36 hours if its been an INSANE week lol.... what the fk?!?!
I think we should all cancel our subscriptions, even if you intend to keep using it? cancel and resub, send them a message measured in statistics and money for them... Im personally going over to codex...

Same issue here and just had the most frustrating experience ever trying to get help using the terrible AI chat on claude.ai

Told me to run '/bug' to report a bug, and '/doctor' to check the integration health - neither of those are commands apparently. The agent did decide that when I said '/bug' I meant '/debug' and started to do something, I immediately cancelled and saw that it had used another 1% of the session usage.

The agent told me to come here and log an issue. There are already a bunch and this was the oldest one I could find so adding mine here.

Details:

• Fresh session, one prompt which had used 74% of context when it completed, used up 18% of my session usage
• Cancelled "/debug" prompt increased usage by 1%
• Working in a VS code workspace containing ~200 files
• No MCP or anything else running (but the context says it's using tokens for gmail and google calendar MCP tools, even though I don't have those connected??)

My single prompt:

**do not make any changes.**

**do not attempt any database operations (npx wrangler d1... commands) - database state has been manually verified.

check whether migration stages 1 - 5 have been fully implemented according to the prompts in the doc/ folder.  Prompts are v2-02-migration-stage-1-plan.md through v2-06-migration-stage-5-plan.md.  Report all gaps between the implementation and the expected state based on the prompts being fully implemented.

Output from /context after the single prompt (and cancelled /debug):

Context Usage

Model: claude-sonnet-4-6
Tokens: 114.3k / 200k (57%)

Estimated usage by category

Category | Tokens | Percentage
-- | -- | --
System prompt | 6.5k | 3.2%
System tools | 8.2k | 4.1%
MCP tools (deferred) | 303 | 0.2%
System tools (deferred) | 10.1k | 5.0%
Memory files | 1.6k | 0.8%
Skills | 712 | 0.4%
Messages | 97.6k | 48.8%
Free space | 52.5k | 26.2%
Autocompact buffer | 33k | 16.5%

MCP Tools

Tool | Server | Tokens
-- | -- | --
mcp__claude_ai_Gmail__authenticate | claude_ai_Gmail | 148
mcp__claude_ai_Google_Calendar__authenticate | claude_ai_Google_Calendar | 155

Memory Files

Type | Path | Tokens
-- | -- | --
Project | /xxxxxxxx/.claude/CLAUDE.md | 1.5k
AutoMem | /xxxxxxx/.claude/projects/xxxxxxx/memory/MEMORY.md | 77

Skills

Skill | Source | Tokens
-- | -- | --
update-config | undefined | 173
keybindings-help | undefined | 61
simplify | undefined | 23
loop | undefined | 77
schedule | undefined | 66
claude-api | undefined | 76
grill-me | Project | 60
test-formulas | Project | 7
todo | Project | 15
reset-test-sheet | Project | 22
todo:remove | Project | 26
todo:list | Project | 20
todo:done | Project | 15
code:review | Project | 15
design:review | Project | 29
design:sync | Project | 27

Same issue here
MacOS CLI v2.1.87

I'm experiencing the exact same issue. Adding my data points for visibility.

Session comparison on the same project, same model (Opus 4.6), same CLI (v2.1.88):

• March 30: 8 hours active work, 42 agent invocations, 28 files, ~2420 lines, 14 commits → stayed under quota
• March 31: 30 minutes active work, 18 agent invocations, 9 files, ~300 lines, 6 commits → 20% quota consumed in 10 minutes

Less than 1/10th of the workload consuming quota at 10x the rate. Nothing changed on my end.

What makes it worse: a significant portion of the quota is burned on correction passes — fixing code that Claude Code itself produced incorrectly. Paying twice for wrong output.

Session IDs:

• March 30 (normal): f0b1ed07-9046-4621-84bd-788eda376aec
• March 31 (abnormal): dfd389a4-57cf-414c-8070-bbe7393dff89

Filed as #41662 (duplicate of this one). 4-month paying customer, $100/month Max plan.

+1. Max 20, v2.1.89, April 1. 100% in ~70 min after reset. Light usage, no bulk ops. Never happened before.

Full report: #41788
Related: #38335, #38239, #40790, #6457, #40895, #41055, #41174, #41550, #41617, #41663, #41779, #41802

I've been experiencing the same issue on Max 20 ($200/mo) — rate limit 100% exhausted in ~70 minutes.

After setting up a monitoring proxy using the official ANTHROPIC_BASE_URL env var, I identified two cache bugs as the root cause (#40524, #34629) and measured the impact: cache read ratio dropped to 4.3%, meaning ~20x token inflation per turn. After applying workarounds it stabilized at 89-99%.

Full analysis with per-request measured data, safe workarounds, and community references (including cc-cache-fix): https://github.com/ArkNill/claude-code-cache-analysis

<img width="583" height="360" alt="Image" src="https://github.com/user-attachments/assets/2cea7f11-1d83-4a8b-8301-7f9456da7905" />

Done in ~3 days and i slowed down the last day and i was on .88. and .89...
Zero difference for me... Thats about HALF my usage this week...

This is max20... 200 usd per month... Unacceptable....

Everyone should cancel their sub, even if you intend to resume it in x weeks? Cancel it... If 10-100k of us do it? It will be statistically big enough to force their hand... Its the only way at this point...

Follow-up — precautions update (April 2, 2026)

Additional findings since my root cause analysis. These behaviors compound the cache bug to drain quota even faster:

Avoid: --resume (token bomb), /dream & /insights (background drain), v2.1.89 (cache bug + terminal regression)

Careful with: sub-agents (Haiku = 0% cache read, 317K/31 calls), multiple terminals (parallel drain), large CLAUDE.md

Recommended: v2.1.81 fixed + fresh sessions + proxy monitoring (ANTHROPIC_BASE_URL). Even with cache at 91-99%, drain is faster than 2 weeks ago — likely a server-side change.

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

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

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

This bug seems to be back for me. in 9 min of a brand new session today i use 63% of my usage with one simple prompt. Not sure whats happening this time. I also had extra usage on from the $100 credit I was given from the last issue and it burned through $25 of that credit from the prior session when I didnt realize this issue was happening again. Not sure if anyone else is experiencing something similar again.

Today Claude Code limits burning like devil. I have Premium seat in Team plan and before never happened this kind of thing, it just stopped working aster 3 tasks, before I was able to work whole day without thinking about limits, like I am not high load user, just standard user, the standard even was oki for me, btw today it is insane disgusting to work, it just burned whole limits, with 3 parallel tasks, which even not hard tasks, for example one task just fixed favicon issue in Nextjs lol. And that disgusting chat support, that 100 times tell me that I am not Premium seat user lol. IDK bros we don't need Fable 5 if Claude Code will stop working after 3 tasks with Premium seat. Thank you for awesome and insane work btw, but this can be fixed as you are still best in whole market and each dev knows about it. That's it. Please check what happened as it is impossible to use.