[BUG] Fable 5.1 configured at `high` effort, but model receives `reasoning_effort: 10`

Summary

My session is configured for high effort on Fable 5.1 everywhere I can configure it, but the model itself reports receiving reasoning_effort: 10.

Environment

• Claude Code 2.1.278, local terminal (CLI)
• macOS, Darwin 25.6.0
• Model: claude-fable-5-1[1m]
• Date: 2026-09-21

Configuration

| Source | Value |
|---|---|
| Session env CLAUDE_EFFORT | high |
| ~/.claude/settings.json → modelSettings["claude-fable-5-1"].effortLevel | "high" |
| ~/.claude/settings.json → top-level effortLevel | "medium" |
| Project / local settings | no effort overrides |
| UI | shows "Fable 5.1 high" |

Observed

Asked the model what reasoning-effort value it sees. It reports reasoning_effort: 10, consistently across turns in the same session.

The numeric scale is undocumented, but users commonly report values such as 50 and 85, which implies a 0–100 scale. On that scale 10 is near the bottom, not "high".

Expected

A value consistent with high, and documentation of how effort labels map to the number the model receives.

Impact

I spent over 14 hours in Claude Code on Fable today and completed zero work. The entire day went to re-explaining basic context to the model and correcting trivial mistakes it made repeatedly. An effort value of 10/100 is the most complete explanation I have for that.

Over the same day my Fable quota and overall weekly quota were dramatically consumed.

I was charged for high effort and shown "high" in the UI while the model appears to have been run at a small fraction of that. If this is confirmed, I consider it fraud: paying for one service level, being told I'm getting it, and receiving another. I'd like:

1. Confirmation of what effort value my requests were actually sent with today.
2. The label → number mapping documented.
3. My usage for today reviewed.

Repro

1. Set modelSettings["claude-fable-5-1"].effortLevel to "high" in ~/.claude/settings.json.
2. Start a session on claude-fable-5-1[1m].
3. Ask the model what reasoning-effort value it sees.

Possibly related

• #88842 — effortLevel: "high" in settings.json but requests are sent at low
• #89080 — Effort picker shows a level as selected while requests are silently applied at a lower level