Claude Code's /advisor puts Opus on call, and its tokens still count

My take: The split makes sense: a cheaper model does the routine, a stronger one weighs in at decision points. Its tokens still count against your plan.
What to know
- In Claude Code,
/advisorpairs your main model with a second, usually stronger one that Claude consults during a task. It tends to call it before choosing an approach, when an error keeps coming back and before saying the work is done, but Claude decides when: the timing is "model-driven rather than rule-based", says the documentation. The advisor reads the whole conversation and sends back guidance. The feature is labeled experimental. - Three ways in:
/advisor opusduring a session (saved as your default), theadvisorModelsetting, orclaude --advisor opusfor one session./advisor offstops it. The advisor must rank at or above the main model: with Opus 5.5, only Opus 5 or later and Fable qualify. - Each call bills the advisor's tokens on top of the main model's. On a subscription, they count toward your plan's limits, and a Fable advisor bills to usage credits on plans where Fable does. In Anthropic's own test, published on April 9, 2026, Sonnet 4.6 with an Opus 4.6 advisor scored 2.7 points higher than Sonnet alone on SWE-bench Multilingual, a coding benchmark, and cost 11.9% less per task.
Spotted on @henriexploria, in French.
What changes
The reel promises you will never hit your limit again. The saving comes from a cheaper main model with a strong one on call, against the strong one all the time. If you already work on Opus 5.5, an Opus advisor adds tokens. A Fable advisor, though, is a middle path for the hard task, short of switching to Fable 5.2.
Try it on a long job, a refactor or a bug that keeps coming back. Note /usage before and after, and press Ctrl+O on the "Advising" line to read what the advisor said. Turning it on or off keeps your prompt cache.
Source: Anthropic, “The advisor strategy: Give agents an intelligence boost”, April 9, 2026
Drafted with Claude, Anthropic's AI, from my take. Reviewed and owned by me.