[BUG] Model silently switches to Fable 5 (consuming its limit) while /model still shows Opus/Sonnet; switching models does not clear the block

Open 💬 3 comments Opened Jul 5, 2026 by pj0201

Preflight Checklist

  • [x] I have searched existing issues 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?

While using Opus/Sonnet (explicitly selected via /model), the model silently switched to Fable 5 in the background without any user action, and Fable 5 usage credits were consumed. This happened repeatedly — documented ~13 switches within a single ~1 hour window in one session.

Once the Fable 5 usage limit was hit, running /model to switch to Sonnet 5 or Opus 4.8 did NOT clear the block. Both the /model output and the on-screen model indicator kept showing Opus/Sonnet the whole time, yet the usage indicator (% remaining) kept draining as if Fable 5 was the one actually running, and the assistant kept returning "You've reached your Fable 5 limit" regardless of the model shown as selected.

I contacted Anthropic support about the billing impact of this separately; one of the affected charges has already been refunded, which suggests the underlying issue is acknowledged as legitimate on the billing side, even though the root cause (silent model switching) has not been addressed here.

Related issues with similar symptoms: #74245, #66884, #74000, #73692. This report adds a different failure signature — repeated bidirectional oscillation (not a single one-directional fallback) plus a real-time screenshot + raw session-transcript correlation as evidence.

What Should Happen?

The model actually running should always match what /model and the on-screen indicator display. If Fable 5 is used in the background for any reason (safety fallback or otherwise), that should be reflected transparently in the model indicator and usage tracking — not silently attributed to a different displayed model. Additionally, once blocked by a Fable 5 limit, /model should be able to actually restore a usable session on the newly selected model.

Error Messages/Logs

You've reached your Fable 5 limit. Run /usage-credits to continue or switch models with /model.

Steps to Reproduce

  1. Set model to Opus 4.8 or Sonnet 5 via /model
  2. Continue a normal working session over an extended period (this occurred during a long multi-hour session with heavy tool/subagent usage)
  3. Observe (via the local session transcript at ~/.claude/projects/<project>/<session-id>.jsonl, field message.model) that the model silently switches to claude-fable-5 for stretches of time, alternating back and forth with claude-opus-4-8, without any /model command being run
  4. Eventually the session hits "You've reached your Fable 5 limit"
  5. Run /model to switch to Sonnet 5 → limit message persists
  6. Run /model to switch to Opus 4.8 → limit message still persists

Note: I cannot yet reproduce this in a minimal isolated repro; it was observed during heavy real-world usage. I can provide the redacted raw JSONL segment and a real-time screenshot showing steps 4-6 on request.

Claude Model

Not sure / Multiple models

Is this a regression?

I don't know

Last Working Version

_No response_

Claude Code Version

2.1.201 (Claude Code)

Platform

Anthropic API

Operating System

Ubuntu/Debian Linux

Terminal/Shell

WSL (Windows Subsystem for Linux)

Additional Information

I have two pieces of supporting evidence available on request:

  1. A real-time terminal screenshot (timestamped 2026-07-03 19:46 JST) showing me typing "it somehow switched to FABLE on its own without me noticing?!" immediately followed by /model switches to Sonnet 5 and Opus 4.8, both of which still showed "You've reached your Fable 5 limit."
  2. The raw (redacted) session JSONL excerpt that produced that exact screenshot moment, plus an aggregated timeline of ~13 model-field switches between claude-opus-4-8 and claude-fable-5 within roughly a 4.5 hour window on 2026-07-03.

I did not attach these directly here to keep the report focused; happy to add them as a follow-up comment if useful.

<img width="335" height="418" alt="Image" src="https://github.com/user-attachments/assets/80e2d7f2-a0bf-452f-a0c4-6d3df4143d81" />

View original on GitHub ↗

This issue has 3 comments on GitHub. Read the full discussion on GitHub ↗