[FEATURE] Max plan usage limits silently reduced since Claude 4.6 release
Open 💬 26 comments Opened Feb 26, 2026 by anbeckham
Preflight Checklist
- [x] I have searched existing requests and this feature hasn't been requested yet
- [x] This is a single feature request (not multiple features)
Problem Statement
Problem
Since the Claude 4.6 release, I've been burning through my weekly usage cap dramatically faster on the Max plan ($200/month). This is not increased token consumption on my end — I audited my entire tooling stack looking for rogue API calls or background token drain. There are none. The caps themselves were lowered without a word.
There is zero transparency around this:
- The actual limits are never published — just a vague percentage with no denominator
- No notification when they change — you find out by slamming into a wall mid-week
- No changelog, no announcement, nothing — paying customers are left guessing
Impact
I'm paying $200/month. That's not a casual subscription. Silently pulling back what that buys is not okay.
- I wasted hours investigating whether my own tooling was causing excess usage — it wasn't. It was a quiet cap reduction.
- I can't scope my work for the week because I don't know what "100%" means anymore
- I can't trust that the workflow I rely on today will still be viable next week
- Every major model release apparently comes with a stealth nerf to what Max subscribers get
Proposed Solution
- Publish concrete rate limits for each plan tier. Stop hiding behind percentages.
- Notify users when limits change — email, dashboard banner, anything. Silence is disrespectful.
- Show actual usage numbers, not just percentages — "75% of ?" is useless
- Changelog for limit adjustments — especially after model releases when costs change
- If you need to reduce limits, say so upfront — don't let paying customers discover it by getting cut off
Alternative Solutions
_No response_
Priority
High - Significant impact on productivity
Feature Category
Documentation
Use Case Example
_No response_
Additional Context
_No response_
26 Comments
👏 👏 👏
Seeing the same thing here over the last few days. Something seems to have changed.
I see the same on MAX 100$/month tariff. Within 15 min session and 10 short messages I got 40% of my 5 hour limit used. Before that I have never ever hit limits with my usage patterns.
Same here. I'm on $100 at the moment, but it feels like $20 which I tried recently.
Usually my session limit was 20-30% max, but now it ends after 2 hours. It's impossible to work like that.
Same issue here. Max 5x plan ($100/month), using Claude Code with
claude-opus-4-6.I tracked my token usage across sessions over the past 10 days. Before the 4.6 release I could work intensively all day (2000-4000 messages, 200-350M tokens/day) without ever hitting the limit. Now I'm hitting the cap with significantly less usage (~1000 messages, ~85M tokens).
Daily usage comparison from my session logs:
| Date | Total tokens | Messages | Hit limit? |
|------|-------------|----------|------------|
| Feb 16 | ~345M | 4,132 | No |
| Feb 20 | ~228M | 2,853 | No |
| Feb 23 | ~196M | 2,020 | No |
| Feb 26 | ~85M | ~983 | Yes |
The limits were clearly reduced without any notice. Transparency would be greatly appreciated.
@thomasmilian2 Thanks for the idea. In my logs I see daily usage up to 174M tokens without hitting any limits. For today the total is 83M. And I hit the session limit twice already!
@Leksat I can confirm I've encountered similar issues. I've already hit the session usage limit twice today. Currently, I'm at 59% usage after only an hour of work. The usage experience was much better before.
Hm. Interesting:
So it feels like the limit was reduced by 3 times.
If Anthropic had reduced the limits for all subscribers, they would likely have done it gradually, little by little. Given that we see only a small number of reports on this issue, I suspect that we might be the lucky ones selected for the A/B test in the "what will happen if we reduce the limits dramatically" category.
Just an observation.
For me I only see this when using Opus (doesn't matter 4.5 or 4.6).
With Sonnet I am still able to make as many requests as I used to do with Opus before today without any noticeable % of usage increase.
@nmtitov Thanks for the suggestion, I have been burning through my usage today which lead to finding this issue. Drastic reduction in limits. Switching to Sonnet for majority of my asks has helped reduce the rate at which I am gaining % usage. Hadn't hit a limit in over 6 weeks and now I'm slowing down in order to preserve usage.
Same for me. No change in how I'm using claude. I see usage increase without doing anything. This is very annoying. Let's have truth in advertising.
I'm on a Team plan with a Premium seat. Starting February 26, my usage limit consumption has increased dramatically without any change on my end.
Before Feb 26: Normal usage throughout the day, rarely hitting limits.
Since Feb 26: Reaching 100% within a short session. Today, 6% was consumed by just ~10 light conversational prompts (no coding, no heavy tasks) using Opus 4.6 via Claude Code.
same issue, feels like I am on Pro account again, exactly the same limits (max 100$ plan)
Confirming. I switched to Sonnet, and the limit usage feels the same as before. So only Opus is affected.
Confirming to. Launched a new session this morning, said "Hello" and "Just answer this message." => 6% on the current session ...
I'm using Sonnet and it seems better. But I also feel like he's using it too. But is Sonnet 4.6 comparable to Opus 4.5?
Can confirm. Have a Pro Plan and today immediately hit my rate limit after roughly 200k tokens with opus.
Never (even yesterday) hit the 5 hour rate limit before with much longer requests.
I have not even used it for coding but only for my PKM.
Definitely unusable like this.
Yes it's true, Claude lowered limits on PRO MAX plans without official announcement.
This issue is already fixed here https://github.com/anthropics/claude-code/issues/29000#issuecomment-3975661919
its the claude code version not the model. model .52 and below seem unaffected by this.
The hotfix in v2.1.62 did not resolve the issue. I also downgraded to v2.1.52 (pre-bug version) and the consumption rate remains the same.
Since the issue persists regardless of client version, this appears to be a server-side problem.
Agree on the transparency point, but there's a practical angle here worth considering even if Anthropic publishes exact limits tomorrow.
The root issue is that Claude Code has no awareness of what code is actually relevant to your task. It reads entire files into context, often pulling 8-10x more code than it needs. When per-token cost goes up (whether from a hungrier model or lower caps), that waste is what kills your quota.
I ran the numbers on my own project: ~18k tokens per query, ~2.4k actually needed. That's 85% waste on every single request. Multiply that across a workday and it explains why you're hitting walls mid-week.
The fix that worked for me: serve the agent a pre-computed dependency graph via MCP instead of raw files. Agent gets only the functions, classes, and connections relevant to the task. Cuts ~70% off input tokens, which effectively gives you 3x the quota without Anthropic changing anything.
I built vexp.dev for this but the principle is simple, anything that stops the agent from reading your entire codebase on every prompt will buy you significantly more runway per week.
Max 20x subscriber — $200/month, the highest individual tier Anthropic offers. I chose it to get work done without limit anxiety, period.
Since the Opus 4.6 release, my weekly quota evaporates in a fraction of the time. Same workflows, same tooling, nothing changed on my end. What changed is that Anthropic silently reduced what $200/month buys — zero announcement, zero changelog, zero transparency.
If 4.6 costs more compute — say that. Publish new limits. Adjust pricing. Let people make informed decisions. What you don't do is charge the same while delivering less and hope nobody notices. That's not a pricing strategy, that's deception.
$200/month is not a casual subscription. People on this tier are professionals scoping real work around predictable capacity. When "100%" silently means 60% of last month, nothing is plannable.
Here's the part Anthropic should really think about: this approach specifically targets your most loyal, highest-paying users — the ones who chose to pay the absolute maximum because they believe in the product — and systematically converts them into vocal critics. Not because the model got worse. Not because of a price increase. Because you chose silence over honesty. You are manufacturing your own haters out of your biggest advocates.
Publish concrete, versioned limits. Announce changes before they hit. The current strategy of quiet downgrades erodes trust faster than any price hike ever could.
Since yesterday, it seems like the limit has been lowered even more, now I'm hitting the limit pretty much all the time.
Are you using latest version now? /status and version will show you that info. It was patched or atleast improved the day after this thread started. I'm on 2.1.69.
If you happen to be on a slightly older version. /exit and then claude --continue will bring you back to where you left off on latest.
Yup, these changes are drastically changing my workflow after I've come to rely on Claude Code quite a lot. I am now buying extra usage as an emergency measure but a single query can easily take up 5% of that extra usage as well. This is a trust issue - I hope to soon see this rectified cause I really like loving Anthropic!
Noticed this too.
I'm usually very mindful of context and try to avoid heavy requests, so on the Pro plan I barely felt the limits at all. Now Max 5x isn't enough -- just typing "hi" in a Claude chat with Opus 4.6, no extended thinking, burned 1% of the 5-hour limit. So the limit is literally 100 "hi" messages, each in a fresh chat?
Either the system prompt weighs a ton, or they're cutting limits whenever they feel like it.
Really disappointing -- Claude is genuinely an amazing model, but this is no way to treat paying customers.