Fable used as the advisor does not appear on the Fable weekly usage meter (Max 5x)

Summary

Fable used as the advisor does not appear on the weekly Fable usage meter, while the same model used as the main model does. I could not determine whether advisor usage lands on the all-models meter instead. Filing this as a clarification request rather than a bug report — I cannot tell from the docs whether this is intended, and it decides whether pairing a Fable advisor is a sensible cost strategy.

Advisor spend is currently invisible in the UI, so there is no way for a user to answer this for themselves without parsing session transcripts.

Environment

• Claude Code 2.1.260 (desktop app), Windows 11
• Max 5x, usage credits off
• Main model Opus 5; advisorModel: "claude-fable-5-1" in a project's .claude/settings.local.json

Evidence 1: a window with advisor-only Fable usage

Between 2026-09-05 01:02 (when advisorModel was first set) and 2026-09-06 00:43, Fable was never used as a main model. In that window the session transcripts record 27 advisor calls as usage.iterations[] entries of {"type":"advisor_message","model":"claude-fable-5-1"}, totalling 8,203,783 input and 124,804 output tokens.

At the end of that window, Settings → Usage showed:

すべてのモデル (all models)  4% used
Fable                       0% used  —  "Fable はまだ使用していません" (Fable has not been used yet)
使用クレジット (credits)     $0.00

The Fable row did not read a rounded-down small value; it asserted that Fable had not been used at all.

From 2026-09-06 00:43 the same account began using Fable as a main model (346 assistant messages, ~1.36M output tokens). The Fable weekly meter now reads 11%. So the meter does move for main-model usage, on a token volume far smaller than the advisor traffic that moved it not at all.

Evidence 2: controlled before/after

/usage immediately before and after a run that forced 5 advisor consultations over a large transcript:

| | all models | Fable |
| --- | --- | --- |
| before | 30% | 11% |
| after | 31% | 11% |

That run consumed, per the transcript:

• advisor (Fable 5.1): 5 calls, 1,078,594 input / 24,691 output
• executor (Opus 5): 139,680 output, 21,600,446 cache read, 1,057,740 cache creation

The Fable meter did not move. The all-models meter moved 1pp, but the executor's own consumption in the same run is large enough to account for that on its own, so this test cannot attribute the 1pp to advisor usage — see "What I could not determine".

What the docs say

• Claude Code advisor page: "On subscription plans, advisor usage counts toward your plan's usage limits, except that a Fable advisor bills to usage credits on plans where Fable usage does." On Max, Fable does not bill to credits, so the exception should not apply — consistent with credits staying at $0.00.
• Fable models on your plan: on Max, Fable draws from the same weekly pool, capped at 50%. No mention of advisor or server tools.
• API advisor page: "Top-level usage fields reflect executor tokens only. Advisor tokens are not rolled into the top-level totals because they are billed at a different rate."

That last sentence is a plausible mechanism: if the subscription meters aggregate top-level usage, advisor tokens would be invisible to them by construction. But nothing states that the subscription meters are built that way.

What I could not determine

Whether advisor tokens contribute to the all-models meter. In Evidence 2 the executor consumed roughly two thirds of the run's cost, so a 1pp move is explained without the advisor. Distinguishing the two needs either a much larger advisor-dominated burn or finer-grained usage data than percentages.

Questions

1. Is advisor-model usage intended to be excluded from the Fable weekly meter?
2. If advisor usage draws on the plan pool as the Claude Code docs say, which meter is expected to reflect it?
3. If it is excluded from both, is that intended, and does the 50% Fable cap apply to advisor usage at all?
4. Could /usage report advisor spend separately? The data already exists in usage.iterations[]; today a user has to parse ~/.claude/projects/**/*.jsonl to see what a Fable advisor cost them.

Side observation

The desktop app's Settings → Usage panel and /usage disagreed materially at nearly the same moment — the panel showed 4% for all models while /usage reported 30%, with the same reset time. The panel displayed "最終更新: たった今" (updated just now). If that panel is cached more aggressively than its label suggests, it is worth separating from the question above, since it can make any usage investigation misleading.