[BUG] Sonnet model consuming both "All models" AND "Somente Sonnet" limits
Preflight Checklist
- [x] I have searched existing issues and this hasn't been reported yet
- [x] This is a single bug report (please file separate reports for different bugs)
- [x] I am using the latest version of Claude Code
What's Wrong?
Bug: Sonnet model consuming both "All models" AND "Somente Sonnet" limits
Description
When using only Sonnet model in Claude Code CLI, both usage limits are being consumed:
- "All models" limit (which should exclude Sonnet)
- "Somente Sonnet" limit
This appears to be incorrect billing/tracking behavior.
Expected Behavior
Using Sonnet should only consume the "Somente Sonnet" quota, not the "All models" quota.
Actual Behavior
Both quotas are being consumed simultaneously when using Sonnet.
Evidence
| Limit | Usage | Reset |
|-------|-------|-------|
| All models | 37% | Monday 18:00 |
| Somente Sonnet | 22% | Tuesday 09:00 |
I have only been using Sonnet model, yet "All models" shows higher consumption than "Somente Sonnet".
Environment
- Product: Claude Code CLI
- Plan: Claude Max
- Model used: Sonnet only
- OS: Windows/WSL2
Steps to Reproduce
- Use Claude Code CLI with default Sonnet model
- Check usage limits in account dashboard
- Observe both "All models" and "Somente Sonnet" being consumed
Additional Context
Screenshot attached showing the dual consumption issue.
<img width="640" height="426" alt="Image" src="https://github.com/user-attachments/assets/a676732f-2c0f-4250-917b-eb4f90fb0e7e" />
<img width="1098" height="267" alt="Image" src="https://github.com/user-attachments/assets/bd031e37-4fcb-40ce-b394-18d6fd5997ed" />
What Should Happen?
Using Sonnet model should ONLY consume the "Somente Sonnet" quota, not the "All models" quota. These should be separate limits.
Error Messages/Logs
Steps to Reproduce
- Open Claude Code CLI (v2.0.71)
- Use only Sonnet model (default) for any task
- Check usage limits in account dashboard at https://console.anthropic.com
- Observe that BOTH "All models" AND "Somente Sonnet" limits are being consumed
- Expected: Only "Somente Sonnet" should be consumed when using Sonnet
Claude Model
None
Is this a regression?
Yes, this worked in a previous version
Last Working Version
_No response_
Claude Code Version
v2.0.71 - Sonnet 4.5
Platform
Anthropic API
Operating System
Windows
Terminal/Shell
Terminal.app (macOS)
Additional Information
_No response_
15 Comments
This issue has been inactive for 30 days. If the issue is still occurring, please comment to let us know. Otherwise, this issue will be automatically closed in 30 days for housekeeping purposes.
Still experiencing this issue. Using Sonnet drains both the "All models" and "Sonnet only" quotas simultaneously. Would love to see this resolved - it defeats the purpose of having a separate Sonnet allowance.
i am experiencing the same issue.. which concerns me if i max out all model and sonnet has still credits will i still be able to use sonnet or not?
I'm a Claude Max subscriber and I'm experiencing the exact same issue. Using only Sonnet, both the "All models" and "Only Sonnet" quotas are being consumed simultaneously.
This has been open since December 2025 — over 3 months — with no fix or official response from Anthropic. For a Max plan subscriber paying $100-$200/month, this is unacceptable. The "Only Sonnet" limit was introduced specifically to give users more headroom, but if it drains the "All models" quota at the same time, the feature is essentially broken.
Please prioritize this fix or at least provide an official clarification on whether this behavior is intentional or not.
Same issue (max 200). I agree, that's unacceptable
Any progress on this issue?
I have same question.
I have same question.
any news ???
I have same question.
Any official answer on this? Has these two independent meters been explained anywhere?
According to the (current) tooltip, it says: "Your Sonnet usage counts toward this limit, as well as your weekly and 5-hour session limits". This is intentional behavior. Sonnet has it's own limit, and it is not separate from other limits - it's in addition to.
I'm just guessing here, but I imagine Sonnet is the most-used model and they had to put a separate limit on it to save their compute. They are really struggling with compute demand lately.
Any update?
I want to add that there is no mention of this on the upgrade screen and downgrading immediately does not grant a refund.
Can confirm this at the API level. I built a local proxy that intercepts all Claude Code traffic and logs every
anthropic-ratelimit-*response header. Tested with both Opus 4.6 and Sonnet 4.6 requests — identical headers returned for both models, all prefixedanthropic-ratelimit-unified-*. There are no per-model headers in the API responses at all.If a separate Sonnet pool existed on the backend, there would be separate headers. There aren't. The UI bar appears to be tracking something client-side that doesn't match server reality.
Proxy is open source if anyone wants to verify: https://github.com/InertiaUK/claude-quota-proxy