[BUG] Same token cost, 6-12x difference in weekly quota consumed (Fable 5.1, Max 20x)

### Preflight Checklist

- [x] I have searched [existing issues](https://github.com/anthropics/claude-code/issues?q=is%3Aissue%20state%3Aopen%20label%3Abug) 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?

## Quota consumption disproportionate to token usage (Fable 5.1, Max 20x)

Two Claude Code sessions, same model, same plan, 2026-09-04:

| | Session A | Session B |
|---|---|---|
| Tokens | 221.2k in / 235.4k out / 513M cache read / 6.2M cache write | 129.3k in / 116.4k out / 550M cache read / 4.3M cache write |
| Weighted cost | ~60.5M | ~61.1M |
| Weekly quota consumed | ~60% | 5-10% |

Weighted cost = input-equivalents (cache read 0.1x, cache write 1.25x, output 5x input).

**Problem:** Identical token cost within 1%. Quota consumption differs ~6-12x.
Session B is marginally more expensive by token weight yet consumed far less quota.

**Ruled out:** counter lag (post-session readings), bucket mismatch (both `weekly_scoped`), model mismatch (both Fable 5.1).

### What Should Happen?

**Expected:** Quota consumption scales with weighted token cost. Two sessions within 1% on tokens should consume comparable quota.

**Actual:** ~6-12x difference.

### Error Messages/Logs

```shell

```

### Steps to Reproduce

**Steps to reproduce:**
1. Run a long agentic Claude Code session on Fable 5.1 (Max 20x), high cache-read volume.
2. Record `/usage` after the session settles.
3. Run a second session with comparable token profile.
4. Record `/usage` again.
5. Compare quota deltas against weighted token cost — they diverge ~6-12x.

### Claude Model

Other

### Is this a regression?

Yes, this worked in a previous version

### Last Working Version

_No response_

### Claude Code Version

2.1.260 (Claude Code)

### Platform

Anthropic API

### Operating System

Windows

### Terminal/Shell

PowerShell

### Additional Information

_No response_

No repository file or test is identified in the report. Start by reproducing the two long Claude Code 2.1.260 sessions from Windows PowerShell, record /usage after each settles, and compare quota deltas with the reported weighted token costs. Done means comparable token profiles produce comparable weekly quota consumption.

由索引模型根据 Issue 内容生成。