/effort xhigh silently switches model from Fable 5 to Opus 4.8, and reverting effort does not restore it

Status Open
Reported on v2.1.215
Maintainer reply None cached
Activity 0 comments · opened Jul 20, 2026

What happened

Setting /effort xhigh silently switched the session model from Fable 5 to Opus 4.8, mid-session, with no notice. Lowering effort back to high did not restore Fable 5 — the session stayed on Opus 4.8 until I manually ran /model.

This is not quota-related (see "Not quota" below).

Evidence (from the session transcript)

Model per assistant message, extracted from ~/.claude/projects/<proj>/<session>.jsonl:

2026-07-18T21:42:15Z  claude-fable-5     <- session start, model explicitly selected as Fable 5
...                   claude-fable-5     <- stays Fable 5 for ~2 days of work
2026-07-20T20:47:25Z  (user runs /effort -> "Set effort level to xhigh")
2026-07-20T20:47:44Z  claude-opus-4-8    <- next assistant turn, 19s later: model changed
...                   claude-opus-4-8
2026-07-20T20:49:42Z  (user runs /effort -> "Set effort level to high")
...                   claude-opus-4-8    <- NOT restored; stays on Opus
2026-07-20T20:5xZ     (user runs /model -> Fable 5)  <- only this restored it

The /effort picker described xhigh as: Deeper reasoning than high, just below maximum (Fable 5, Opus 4.7+, Sonnet 5) — i.e. it lists Fable 5 as supporting xhigh, so a model change should not have been necessary.

Not quota

At the moment of the switch, /usage showed ample headroom:

  • Current session: 7% used
  • Current week (all models): 5% used
  • Current week (Fable): 9% used
  • Usage credits: off

So this was not credit exhaustion or a weekly-limit fallback.

Impact

The user is running Fable 5 deliberately as the orchestrator for a long-lived multi-agent session. A silent downgrade mid-session:

  1. changes model behaviour with no visible signal in the conversation,
  2. is not reverted when the triggering setting is reverted,
  3. led the user to conclude the model itself was underperforming, because nothing indicated the model had changed.

At minimum, an effort change that forces a model change should say so explicitly, and reverting the effort should revert the model.

Expected

  • /effort xhigh on Fable 5 should keep Fable 5 (the picker lists Fable 5 as supported), or
  • if a model change is genuinely required, print an explicit notice (e.g. "xhigh is not available on Fable 5 — switching to Opus 4.8"), and restore the previous model when effort is lowered again.

Environment

  • Claude Code 2.1.215, macOS (darwin 25.5.0)
  • Selected model: Fable 5 (claude-fable-5)
  • Observed fallback model: claude-opus-4-8

View original on GitHub ↗