Telemetry off in Claude Code? Its Fable row still picks Fable 5.1
If you've turned off Claude Code's telemetry, the Fable row in its /model picker still points to Fable 5.1, the older and pricier model. Pick Fable 5.2 by name instead, with /model claude-fable-5-2 or --model fable.
The 30-second version
With telemetry off, Claude Code 2.1.322’s /model picker offered Fable 5.1 as “Fable”, and picking it sent our request to Fable 5.1.
- Who’s affected. Sessions with
DISABLE_TELEMETRY=1orDO_NOT_TRACK=1, including VS Code when the editor’s telemetry is off. - What it costs. Fable 5.1 is 25% dearer than Fable 5.2 at API prices, and older.
- The fix. Type
/model claude-fable-5-2, or start with--model fable. Both used Fable 5.2 with telemetry off. - For a picker row, add a
modelPickeroption forclaude-fable-5-2to your user settings.
What the picker shows
With telemetry on, our Claude Code 2.1.322 picker listed Fable 5.2. In an otherwise identical session started with DISABLE_TELEMETRY=1, it listed:
| Row | Label | Model |
|---|---|---|
| 1 | Default (recommended) | Opus 5.5 |
| 2 | Opus | Opus 5.5 |
| 3 | Fable | Fable 5.1 |
| 4 | Sonnet | Sonnet 5.5 |
| 5 | Haiku | Haiku 5.5 |
We picked row 3 for that session and sent a prompt; the debug log shows the request went to claude-fable-5-1. Fable 5.1 costs $10 per million input tokens and $50 per million output tokens; Fable 5.2 costs $8 and $40.
Why telemetry changes the picker
Claude Code builds the picker from Anthropic’s model list only when a feature flag says so, and telemetry-off sessions don’t fetch feature flags. The documentation says DISABLE_TELEMETRY “also disables feature-flag fetching”, but its list of affected features doesn’t mention the picker.
Without the flag, the debug log reads [servedCatalog] off (flag), and the picker falls back to its built-in rows plus extra rows from your account. Our account’s extra row was claude-fable-5-1[1m], labeled “Fable”. The log line was the same with DO_NOT_TRACK=1 and CLAUDE_CODE_DISABLE_FEATURE_FLAGS=1, and read off (essential_traffic) with CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC=1.
The VS Code extension sets the variable for you: its 2.1.322 code starts Claude with DISABLE_TELEMETRY=1 whenever VS Code’s own telemetry setting is off.
Three ways to get Fable 5.2
All three worked with DISABLE_TELEMETRY=1 in our tests:
- Name it.
/model claude-fable-5-2switched the session and saved it as the default for new ones.--model fableand--model claude-fable-5-2also used Fable 5.2. - Add a picker row. A
modelPickerentry,{"modelPicker": {"options": [{"model": "claude-fable-5-2", "label": "Fable 5.2"}]}}, passed with--settings, added a sixth row that used Fable 5.2. Claude Code also reads it from user settings, but not from a project’s settings. - Use the cached list. With
CLAUDE_CODE_GB_DISK_CACHE_WHEN_TELEMETRY_OFF=1in the environment, the picker listed Fable 5.2 again, from the model list an earlier session with telemetry on had saved. The variable isn’t documented.
How we measured
- Version and account. Claude Code 2.1.322, installed separately, on one Max 5x account on 12 November 2026.
- Picker. Terminal sessions with
--setting-sources project,localand--debug-file, with and withoutDISABLE_TELEMETRY=1. - Model used. Headless runs (
claude -p --model fable) under each variable, read from Claude Code’s JSON output; for picker choices, the debug log. - VS Code. The extension’s published 2.1.322 code, not a running editor.
What we didn’t test
- Other accounts. The extra rows come from the account; yours may differ.
- The VS Code extension in use, or Windows and Linux.
- Plan usage. We didn’t measure whether Fable 5.1 or 5.2 uses more of a subscription’s limits.
Companion reading
- What Claude Code’s background-agent labels cost — what telemetry-off sessions did to agent labels on 2.1.318.
- Claude Code usage limits — how the 5-hour and weekly limits work.
- Fable 5.2 vs Opus 5.5 in Claude Code — what Fable 5.2 costs per task, in dollars and in plan limits.
Sources
- GitHub — anthropics/claude-code #108726, “DISABLE_TELEMETRY hides Fable 5.2 from the /model picker”, 10 November 2026.
- Anthropic — Claude Code: Environment variables,
DISABLE_TELEMETRYand “Features that need feature-flag fetching.” Retrieved 12 November 2026. - Anthropic — Claude Code changelog, 2.1.319 and 2.1.321 entries.
- Anthropic — Pricing. Retrieved 12 November 2026.
FAQ
Why doesn’t Claude Code’s /model picker show Fable 5.2? With telemetry off, Claude Code 2.1.322 skips the feature flag that lets Anthropic’s model list replace its built-in picker; in our test, its Fable row then pointed to Fable 5.1. The VS Code extension sets DISABLE_TELEMETRY whenever VS Code’s own telemetry is off.
How do I use Fable 5.2 in Claude Code with telemetry off? Type /model claude-fable-5-2, or start Claude Code with --model fable or --model claude-fable-5-2; all three used Fable 5.2 in our tests with telemetry off. To get a picker row, add a modelPicker option for claude-fable-5-2 to your user settings.
Does the ‘fable’ alias mean Fable 5.2 when telemetry is off? Yes, in Claude Code 2.1.322: --model fable used Fable 5.2 with DISABLE_TELEMETRY=1, DO_NOT_TRACK=1, CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC=1 or CLAUDE_CODE_DISABLE_FEATURE_FLAGS=1.
How much more does Fable 5.1 cost than Fable 5.2? 25% more at API prices: $10 per million input tokens and $50 per million output tokens, against $8 and $40 for Fable 5.2. Cache reads cost $0.25 per million tokens on Fable 5.1 and $0.20 on Fable 5.2.
Was this helpful?