You're paying $200/month for Claude Max. You hit "usage limit reached" after 45 minutes. Your account gets disabled. Phone verification fails. You can't work.
3,400+ upvotes and 3,600+ comments across 13 open issues. This is the #1 pain point in Claude Code.
From anthropics/claude-code#16157 (670 upvotes, 1,440 comments):
"My session credit was exhausted in 45 minutes -- almost like someone else was using it. There's simply no way my prompt could have used it like that. The usage meter looked more like a download meter."
From #38335 (361 upvotes, 466 comments):
"Was picking up some tickets. Always /clear after every ticket and was monitoring my usage on my other screen. On the premium plan of my company and suddenly usage shot from 40% to 85%. What the actual fudge."
From #29579 (85 upvotes, 147 comments):
"Rate limit reached despite Max subscription and only 16% usage. The usage meter shows 16% but I'm locked out."
From #5088 (49 upvotes, 160 comments):
"Account disabled after payment for Max 5x plan. I am so upset that I did not even receive any warning or email that I violated something that led to the disablement."
From #37394 (38 upvotes, 74 comments):
"Woke up this morning, did like 2 prompts and BOOM 100% and the reset just happened less than 45 mins ago. There is no way I used it all!!"
From #9424 (115 upvotes, 110 comments):
"Started hitting my Max x5 (~$100/month) limit on Friday. Weekend: Completely blocked. Monday: Hit the limit AGAIN after only ~2 hours of coding. This is making a $100/month subscription practically unusable."
Claude Code's pricing model has a fundamental conflict:
Anthropic sells unlimited-sounding subscriptions ($20-200/month) but enforces unpublished usage caps. The exact limits are not documented. They change without notice. The usage meter is unreliable. And when you hit the invisible wall, you're locked out for hours.
Claude Code now sends the full conversation history with every request. With the 1M token context window, a long coding session means each turn sends hundreds of thousands of tokens -- even if your prompt is one sentence. Users don't see this. They just see their quota vanish.
From #38239 (56 upvotes, 62 comments):
"The 1M context thing would make sense. Bigger window = more tokens per turn even if your actual prompts are the same size, since the full conversation gets sent every time. The idle drain is the weird one though."
Multiple users report quota decreasing even when idle -- suggesting background processes or auto-compaction consuming tokens invisibly.
From #6457 (30 upvotes, 119 comments):
"Just ran a /compact command today... Being on the Max plan, there's just no way I used all of my limit just running /compact."
| What they promise | What happens |
|---|---|
| "Max plan -- highest usage" | Rate limited after 45 minutes |
| "5x more usage" ($100/mo) | Can't get through a morning |
| "20x more usage" ($200/mo) | Account disabled after payment |
| "Usage: 16%" | "You've hit your limit" |
| Phone verification | Code never arrives (747 upvotes) |
RunCode (github.com/BlockRunAI/runcode) takes a different approach: you pay for exactly what you use. No subscriptions. No limits. No accounts.
npm install -g @blockrun/runcode
runcodeTwo commands. You have access to 55+ models. No rate limits. No invisible caps.
RunCode connects to BlockRun, an AI API gateway that accepts USDC micropayments via the x402 protocol:
You → RunCode → BlockRun → Any model (Claude, GPT, Gemini, DeepSeek...)
signs payment 55+ models
with your wallet pay per token
Every request is paid individually. Your wallet has USDC, you can make requests. No USDC, no requests. No ambiguity. No invisible limits. No "usage meter" that lies to you.
| Model | Cost per 1K tokens | Typical coding session |
|---|---|---|
| Claude Sonnet 4.6 | ~$0.009 | ~$0.50-2.00 |
| Claude Opus 4.6 | ~$0.015 | ~$1.00-5.00 |
| GPT-5.4 | ~$0.009 | ~$0.50-2.00 |
| DeepSeek V3 | ~$0.0004 | ~$0.02-0.10 |
| Gemini 2.5 Pro | ~$0.006 | ~$0.30-1.50 |
| GLM-5.1 | $0.001/call | Promo: $0.001 per call |
| NVIDIA Nemotron Ultra | Free | $0.00 |
A typical developer spends $5-15/week using RunCode. Compare that to $200/month for a Max subscription that locks you out after 45 minutes.
You can still use Claude through RunCode. BlockRun routes to Anthropic's Claude API -- same models, same quality. The difference:
- Claude Max: $200/month, invisible limits, account locks, 45-minute sessions
- Claude via RunCode: ~$2/day for heavy use, no limits, no account needed
And you get access to every other model too. Need GPT-5 for reasoning? Gemini for long context? DeepSeek for cheap bulk work? Switch with /model mid-conversation.
| Claude Max 5x | Claude Max 20x | RunCode | |
|---|---|---|---|
| Monthly cost | $100 | $200 | $20-60 (typical) |
| Rate limits | Yes (unpublished) | Yes (unpublished) | No |
| Usage meter accurate | No (#29579) | No | Yes (exact token count) |
| Account can be disabled | Yes (#5088) | Yes | No account needed |
| Phone verification | Required, often broken (#34229) | Required | Not required |
| Models | Claude only | Claude only | 55+ models |
| Token drain visibility | Hidden | Hidden | Every token tracked |
Use RunCode if:
- You've been rate limited on Claude Max
- You can't verify your phone number (#34229 -- 747 upvotes)
- Your account was disabled after paying (#5088)
- You're in a region where Claude is expensive (#17432 -- 268 upvotes)
- You want transparent, predictable pricing
- You want to use models other than Claude
- You're tired of the usage meter lying to you
Keep Claude Max if:
- You never hit rate limits
- You only use Claude and don't need other models
- You prefer subscription pricing
# Install
npm install -g @blockrun/runcode
# Start (free model to test -- no wallet needed)
runcode
/model nemotron
# When ready, fund your wallet with $5 USDC on Base
# That's enough for a week of heavy codingNo account. No phone verification. No invisible limits. Just code.
Links: