[Bug] Opus 5 and Fable 5 depletes usage limits at the same rate. Usage limits decreased to 1/3 of previous limits on 20x max plan without code changes

Status Open
Reported on v2.1.220
Maintainer reply None cached
Activity 4 comments · opened Jul 28, 2026

<img width="705" height="652" alt="Image" src="https://github.com/user-attachments/assets/301cdb8a-4318-4bd0-bba8-da98b6ded7ed" />
<img width="1011" height="281" alt="Image" src="https://github.com/user-attachments/assets/bd62ff3d-147f-4dd5-b1d9-6ab8906f1d09" />

What's Wrong?

Effective usage on the 20x Max plan dropped to roughly one third this week, with no change on my side — same model, same thinking and effort settings, same type of work.

Comparing the two rate-limit windows reported by /usage:

| | 5-hour window consumed | Corresponding weekly window consumed |
|---|---|---|
| Previously | ~10% | 1% |
| Now | ~3% | 1% |

In other words, the weekly allowance is now exhausted after roughly 3x less work than before. Screenshots above.

What Should Happen?

Weekly-window consumption per unit of work should remain stable, or any reduction should be announced in advance. The advertised 20x Max entitlement and the 50% additional-usage extension both imply limits going up, not down.

Error Messages/Logs

[]

No errors are surfaced — the reduction is only visible by comparing /usage output across weeks.

Steps to Reproduce

  1. On a 20x Max plan, set the model to Opus with high thinking effort.
  2. Run a normal working session.
  3. Run /usage and record the 5-hour window and weekly window percentages.
  4. Repeat on a comparable session in a later week and compare the ratio between the two windows.

Observed: the ratio moved from ~10:1 to ~3:1 (5-hour : weekly).

Claude Model

Opus, high thinking effort.

Is this a regression?

Yes. The change occurred without a client update — version 2.1.220 was in use both before and after, which suggests a server-side limit change rather than a code change.

Claude Code Version

2.1.220

Platform

Claude subscription (Max 20x)

Operating System

macOS (darwin)

Terminal

VS Code

Additional Information

Feedback ID: 273b8592-31b2-44aa-a054-44590cc9753f

Could someone confirm whether this reduction is intentional, and if so, where it was communicated? Others in this thread are reporting the same pattern.

View original on GitHub ↗

4 Comments

GrahamCampbell · 1 month ago

I am seeing usage reduced from 4x lots of Fable to week down to 2.66x, like the 50% extra usage has been removed too early.

macblackstuff · 1 month ago
I am seeing usage reduced from 4x lots of Fable to week down to 2.66x, like the 50% extra usage has been removed too early.

I first noticed this around the time the 50% additional-usage extension was announced, though the reduction I am seeing is more severe than that alone would account for.

This is not the first occurrence, and I have reported similar changes previously. Reducing usage limits without prior notice — particularly while increased limits are being advertised — is hard to reconcile. Clarity on the current limits, and on any planned changes, would be appreciated.

Regsorm · 1 month ago

Independent confirmation of the 3:1 ratio you're now seeing — same plan, same client version, different OS.

Environment: Max 20x, Claude Code v2.1.220, Windows, Opus with high effort.

My weekly window reset Friday evening (24 Jul). I was away Fri–Sat with zero usage, so both meters were genuinely at zero when I started work on Sunday 26 Jul. After ~1 hour of active work the meters read:

Session (5hr) 6% / Weekly (7d) 2% — i.e. 3:1, matching your "now" figure exactly.

I have no "before" baseline of my own to compare against, so I can't independently corroborate the 10:1 → 3:1 change itself — only that the current ratio on my account matches what you're reporting after the change.

Two caveats, to keep this honest:

  • Both readings are integer percentages, so rounding puts the true ratio somewhere in the ~2.2:1 to ~4.3:1 range. The point is the order of magnitude, not the exact figure.
  • My workload is ~90% cache_read by token count (31.1M cache_read vs 465 fresh input over that window, deduplicated by message.id). If the weekly pool meters cache reads above the documented 0.1×, that alone would compress the effective weekly allowance for cache-heavy workloads — see #81234, where I posted the full token breakdown. That's a different candidate explanation from the "+50% promotional usage ended early" theory in this thread, and I can't tell from the client side which one applies.
macblackstuff · 1 month ago

Today's testing showed both Opus 5 and Fable 5 depletes usage limits at the same rate.
Fable 5's rate is expected. Opus 5 however should deplete at a slower rate.