Web app: selected model (Fable-tier) not applied — session reports claude-opus-4-8 instead

Status Open
Maintainer reply None cached
Activity 0 comments · opened Aug 12, 2026

What happened

In the Claude Code web app, I select a specific model (a Fable-tier model) and keep it selected. But the running session's own reported model — the model id surfaced in the system context — is a different model (an Opus-tier model, claude-opus-4-8). So the selected model and the model that actually answers / is reported do not match, and I have to re-select the intended model repeatedly.

Steps to reproduce

  1. In the Claude Code web app, open the model picker and select a Fable-tier model.
  2. Continue the session (including across a context compaction / continuation).
  3. Ask the session what model it is, or inspect the system-reported model id.

Expected

The session runs — and reports itself — as the selected model.

Actual

The session reports claude-opus-4-8 (Opus 4.8) even though a Fable-tier model was selected. The selection does not "stick", most reliably after a compaction/continuation, and the reported model disagrees with the picker.

Impact

On a long, phone/web-driven session the mismatch is silent — the user only notices later that the session ran on a different (often more expensive) model than the one they chose. Cost and behaviour both change without a signal.

Related

This looks like the same root cluster as model-preference-not-persisted:

  • #13242 (settings.json model ignored on startup)
  • #39889 (dispatch/mobile ignores the selected model, no way to correct it)
  • #59516 (model not preserved after /compact, closed as duplicate)

Filing this as the web-app manifestation of the same class: the picker's selection and the session's actually-running/reported model disagree. Ask: persist the selected model across compaction/continuation, and surface the currently active model in the web-app UI so a silent swap is at least visible.

View original on GitHub ↗