Omniscio

OmniscioGuidesClaude Code advisor tool

Claude Code advisor tool: a second opinion on demand

The advisor tool is a way to get a stronger model's judgement without running it for every turn: your main model does the routine work, and a second model is consulted at the moments where plan quality decides the outcome.

This page covers when the advisor helps and when it does not, the three ways to enable it, which model pairings are accepted, what you see in the transcript, what it costs, and how it differs from opusplan, subagents and switching models.

Last checked against Anthropic's Claude Code documentation on November 20, 2026. By the Omniscio team. Updated .

What the advisor is

Anthropic describes the advisor tool as letting Claude consult a second, typically stronger model at key moments during a task: before committing to an approach, when stuck on a recurring error, or before declaring a task complete. The advisor receives the full conversation, including every tool call and result, and returns guidance that Claude applies before continuing.

It runs server-side on Anthropic's infrastructure as a server tool, and works for subscription and API-billed accounts alike. You choose the advisor model; Claude decides when to call it.

It is experimental. Anthropic says behaviour, pricing and availability may change, and it is available only through the Anthropic API, Claude Platform on AWS, or Claude Models on Google Cloud (Anthropic managed) — not Amazon Bedrock, Google Cloud's Agent Platform or Microsoft Foundry. On the last two it needs Claude Code v2.1.300 or later.

Sources: the introduction and the note above it in Escalate hard decisions with the advisor tool, read November 20, 2026.

When it is worth it

The advisor fits long, multi-step tasks where most turns are routine but plan quality decides the outcome: large refactors, debugging sessions where an error keeps recurring, and tasks you want independently checked before Claude declares them done.

It adds less value on short tasks with little to plan, or on work where every turn needs the strongest model. For those, Anthropic's advice is to switch the main model instead. The change model guide covers that, and the effort level guide covers how much thinking each turn gets.

Source: "When to use the advisor" in Escalate hard decisions with the advisor tool, read November 20, 2026.

Three ways to turn it on

  • /advisor command. Run it with no argument to open a picker, or pass a model: /advisor opus. Claude Code confirms with Advisor set to and the model name, and saves the choice to advisorModel in your user settings, where it persists across sessions except in the cases Anthropic's advisorModel entry marks as applying to the current session only.
  • The advisorModel setting. A persistent default without opening a session: { "advisorModel": "opus" } in a settings file.
  • The --advisor flag. claude --advisor opus sets the advisor for one session and overrides the saved setting. Anthropic notes the flag is not listed in claude --help.

Each route enables the advisor only for sessions whose main model supports it. After the session starts, Claude Code shows an Advisor Tool (experimental) is on and may use more tokens · /advisor notification.

/advisor also works where there is no terminal picker — in non-interactive mode with -p, in the Agent SDK, in the desktop app and over Remote Control — from v2.1.260 or later. There, run it with no argument to print the current advisor and the aliases it accepts, with a model to set one, or with off.

Four situations make it exit at launch rather than start: the main model does not support the advisor; the requested model cannot act as one at all (Haiku 4.5 can call the advisor but cannot be one); your organization's availableModels allowlist excludes the requested model; or you requested Fable while your account still needs the usage-credits consent. A background session started with --advisor starts without the advisor instead of exiting. The background tasks guide covers that surface.

Sources: "Enable the advisor", "Use the /advisor command" and "Use the --advisor flag" in Escalate hard decisions with the advisor tool, read November 20, 2026.

Which pairings are accepted

Claude Code ranks models by capability for the advisor role, and an advisor must rank at or above your session's main model. That is why the strongest main models accept almost nothing: with Fable 5 as the main model, only a Fable model is accepted, and with Opus 5 or Opus 5.5, only Fable or Opus 5 or later.

For the common cases:

  • Haiku 4.5 or Sonnet 4.6 main: Fable, Opus, Sonnet or Haiku 5.5.
  • Sonnet 5 or Haiku 5.5 main: Fable, Opus 4.7 or later, Sonnet 5 or later, or Haiku 5.5.
  • Sonnet 5.5 main: Fable, Opus 5 or later, or Sonnet 5.5.
  • Opus 5 or Opus 5.5 main: Fable, or Opus 5 or later.

Set the advisor as fable, opus or sonnet, which resolve to Claude Code's built-in default version for each family and advance with releases. A full model ID such as claude-opus-5-5 works too. Subagents inherit the configured advisor and run the same check against their own model.

Three things can quietly make it not apply. An advisor that ranks below the main model is not attached, and both the /advisor output and a notification say so. An advisor your organization's availableModels allowlist excludes is not used. And if the API refuses a pairing Claude Code attached, Claude Code resends that request without the advisor: the conversation continues, you see no error, and no advisor calls happen. Picking a different advisor with /advisor then takes effect after /clear or /compact, and in new sessions.

On some plans, Fable as the advisor needs your one-time consent to bill Fable usage to usage credits. Until you give it, /advisor fable points you to /model fable rather than saving Fable as the advisor.

Sources: "Choose an advisor model" and "Fable advisor and usage credits" in Escalate hard decisions with the advisor tool, read November 20, 2026.

When Claude consults it

Claude decides. It tends to consult before committing to an approach, when an error keeps recurring, and before declaring a task done, but Anthropic is explicit that the timing is model-driven rather than rule-based. There is no setting to cap or force advisor calls; if you want more or fewer consultations, say so in your instructions, for example consult the advisor before you continue.

The transcript shows an Advising line with the advisor model name while the call runs. When the result returns, the line reports one of three outcomes:

  • Reviewed: the advisor reviewed the conversation, and if it returned readable guidance, Ctrl+O shows it.
  • Declined: the line reads Advisor declined to advise on this request, with its reason available through Ctrl+O.
  • Unavailable: the call failed, and the line reads Advisor unavailable (<error_code>).

Claude generally follows the guidance but adapts when its own evidence contradicts a specific claim: if a recommended step fails when tried, or the file contents contradict the advice, Claude surfaces the conflict rather than following it unconditionally.

Sources: "When Claude consults the advisor" and "What you see during a session" in Escalate hard decisions with the advisor tool, read November 20, 2026.

What it costs, and caching

When Claude calls the advisor, the advisor model reads the conversation, so each call consumes tokens at the advisor model's rates on top of your main model's usage. On API billing you pay those rates. On subscription plans the usage counts toward your plan's limits, except that a Fable advisor bills to usage credits on plans where Fable usage does.

Because Claude calls the advisor at decision points rather than on every turn, pairing a faster main model with a stronger advisor typically costs less than running the stronger model throughout. Advisor usage counts toward the session totals /usage shows; the token usage guide covers reading those numbers.

Caching is left intact. Enabling or disabling the advisor mid-session does not invalidate your main model's prompt cache — unlike switching models — and the advisor's returned guidance is cached as part of the transcript on later turns. The advisor model's own read is not cached, so each call processes the full transcript again. The prompt cache guide covers the cache itself.

Sources: "Cost" and "Impact on prompt caching" in Escalate hard decisions with the advisor tool, read November 20, 2026.

Advisor, opusplan or a subagent

Anthropic lists four ways to combine model strengths, and the choice is about when the stronger model runs:

  • The advisor tool runs at decision points mid-task, when Claude calls it for guidance.
  • opusplan runs Opus during plan mode, when your availableModels allow it, then switches to Sonnet for execution. You enter plan mode to trigger it.
  • A subagent with model set runs the stronger model for the entire delegated subtask, when Claude delegates or you invoke it.
  • /model switches the main model from the next request onward.

The distinction worth keeping: the advisor advises and the main model still does the work, while a subagent with a stronger model does the work itself. The subagents guide and the plan mode guide cover those two.

Source: "Compare with related features" in Escalate hard decisions with the advisor tool, read November 20, 2026.

Turn it off

Run /advisor off, or choose No advisor in the picker, to stop using the advisor. To disable the tool entirely, set CLAUDE_CODE_DISABLE_ADVISOR_TOOL=1: the /advisor command becomes unavailable, any configured advisorModel is ignored, and the --advisor flag is accepted but has no effect.

One requirement is easy to miss. Unless you are on Claude Platform on AWS or Claude Models on Google Cloud (Anthropic managed), the advisor needs a feature flag fetched from Anthropic, so a session with a variable that turns fetching off, such as DISABLE_TELEMETRY, does not have it. Through an ANTHROPIC_BASE_URL gateway, availability depends on whether the gateway forwards the request intact.

Sources: "Turn the advisor off" and "Requirements" in Escalate hard decisions with the advisor tool, read November 20, 2026.

Where these answers come from

No advisor call was made from this site's machine to check the page: the feature is experimental and needs a billing account, so every claim is quoted from Anthropic's documentation.

Common questions

Can I make Claude consult the advisor more often?

Only by asking in your instructions. There is no setting to cap or force advisor calls, and the timing is model-driven.

Does the advisor replace my main model's work?

No. The advisor returns guidance and the main model applies it and keeps doing the work. A subagent with a stronger model is the option where the stronger model does the work itself.

Will the advisor see my whole conversation?

Yes, including every tool call and result. That is also why each call costs a full read of the transcript at the advisor model's rates.

What if the advisor and my own evidence disagree?

Claude is documented to surface the conflict rather than follow the advice unconditionally when a recommended step fails or the file contents contradict it.