Paying $200/month for Max 20x? Users report:
- Quota burns 10x faster some hours than others (5.6%/hr vs 59.9%/hr)
- Rate limited at 70% usage while dashboard shows capacity available
- Max plan produces SHORTER output than Pro ($100 vs $20)
- Same workload, same model — wildly inconsistent experience
As one HN user put it:
"Feels like the whole infra is built on a house of cards and badly struggles 70% of the time. I think my $20 OpenAI sub gets me more tokens than Claude's $100."
A Max 20x ($200/month) user instrumented their API calls and discovered massive variance in quota consumption:
| Session | Duration | Burn Rate | Status |
|---|---|---|---|
| Jan 30 18:41 - Jan 31 02:37 | 7.9h | 9.3%/hr | Normal |
| Jan 31 20:28 - Jan 31 22:01 | 1.5h | 56.0%/hr | 🔴 6x faster |
| Feb 01 15:00 - Feb 01 16:40 | 1.7h | 59.9%/hr | 🔴 6x faster |
Same user. Same plan. Same workload type. 10x variance.
Source: GitHub Issue #22435 — "Legal liability claim ready"
Users hit rate limits while their dashboard shows plenty of capacity remaining:
"Claude Code reports I have hit my rate limit even though usage is reported at around 70%"
"I would also add that the chat function still works, only Claude Code is reporting hitting its rate limit."
You're paying for 100%. You're getting cut off at 70%.
Source: GitHub Issue #22441
This one is wild:
"Claude Max plan ($100/month) produces drastically shorter code output than Pro/Plus plan ($20/month) when using the exact same model (Opus 4.5) and the exact same prompt."
- Pro/Plus: Successfully generates ~5,000 lines of code
- Max: Cannot complete even ~1,000 lines — stops mid-response
Pay 5x more, get less output. The user suspects:
- Different
max_tokensparameter server-side - Routing to different (overloaded?) server pools
- "Intentional throttling to manage compute costs on higher-tier plans"
Source: GitHub Issue #22674
From Hacker News:
"I just purchased a third $200 MAX plan and instantly hit rate limits"
Three Max subscriptions = $600/month. Still rate limited instantly.
Source: HN Comment, Jan 26, 2026
Anthropic gave users bonus capacity over the holidays. Then:
"Anthropic bumped limits over the holidays to drive more usage. Which combined with Opus's higher token usage and weird oddities in usage reporting / capping makes me suspect they want to drive people from Pro → Max without admitting it."
Source: HN Comment, Jan 30, 2026
Multiple reports that Pro tier experience has degraded:
"Even when Claude isn't down on status indicator. I get issues with Claude all the time being very slow where it is basically down. I have to kill the request and then restart it. I wonder if it has to do with being on Pro plan. I rarely had this issue when I was on Max."
Source: HN Comment, Feb 3, 2026
"How about how Claude 2.1.x is 'literally unusable' because it frequently completely hangs (requires kill -9) and uses 100% cpu?"
Source: HN Comment linking to GitHub #18532
| Issue Type | Count |
|---|---|
| "Rate limit" issues | 142 |
| "Hit limit" issues | 126 |
| Total open issues | 5,788 |
19 incidents in 14 days = 1.36 incidents per day
For comparison:
- OpenAI: 1-2 incidents per week
- Google Cloud: 2-3 incidents per week
- Anthropic: 9.5 incidents per week
Airlines do it. ISPs do it. Now AI companies do it.
The pattern:
- Sell "5x more usage" and "20x more usage" tiers
- Don't actually have 20x the capacity
- Hope users don't all use their quota simultaneously
- When they do → throttle, rate limit, blame "holiday bonus expiration"
-
Inconsistent burn rates (5.6%/hr to 59.9%/hr) suggest dynamic throttling based on server load
-
Pro users degraded after Max tier launched — resources reallocated to higher-paying customers
-
Rate limits at 70% — system is hitting actual capacity limits before users hit their paid quota
-
Max produces less than Pro — possible routing to overloaded server pools
-
Official incident on Aug 2025 acknowledged "routing to different server pools" as a bug source
"Anthropic might have the best product for coding but good god the experience is awful. Random limits where you know you shouldn't hit them yet, the jankiness of their client, the service being down semi-frequently. Feels like the whole infra is built on a house of cards and badly struggles 70% of the time."
"I think my $20 OpenAI sub gets me more tokens than Claude's $100. I can't wait until Google or OpenAI overtake them."
"I've cancelled my CC subscription, I'm not paying to support anticompetitive behaviour."
"The extremely blunt advice from Claude was that Claude Code was not suitable for serious software development due to usage limits!"
-
What is the actual capacity for each tier? Not "5x more" — actual numbers.
-
Why do burn rates vary 10x for the same user, same workload?
-
Why does Max produce shorter output than Pro with identical prompts?
-
Why are users rate limited at 70% when their dashboard shows capacity?
-
What percentage of subscriptions can actually use their full quota simultaneously?
| What They Sell | What Users Get |
|---|---|
| "5x/20x more usage" | Undefined, variable limits |
| Premium experience | Rate limited at 70% |
| Max tier ($100-200) | Shorter output than Pro ($20) |
| Professional tool | "Not suitable for serious development" |
| 99.41% uptime | 84% calculated from their own data |
- GitHub #22435 — Quota accounting variance (10x)
- GitHub #22441 — Rate limited at 70%
- GitHub #22674 — Max shorter than Pro
- GitHub #18532 — "Literally unusable"
- HN Comment — "House of cards"
- Full Incident Report — 19 incidents in 14 days
Data collected: February 3, 2026