Deep Dive · XiaoHu Explains

Anthropic's official playbook: use Claude Fable 5 as an "advisor," let Claude Sonnet 5 do the work and eat most of the bill

In advisor mode Fable 5 only gives advice; in orchestration mode Fable 5 dispatches tasks — either way, the lower-priced Sonnet 5 handles most of the actual work
TL;DR
  • Anthropic's official developer account, @ClaudeDevs, shared two multi-model collaboration patterns the team uses internally on July 7, 2026: advisor mode and orchestration mode.
  • Advisor mode: the executor model, Claude Sonnet 5, only calls in the advisor model, Claude Fable 5, for critical judgment calls. On SWE-bench Pro, this combo hit about 92% of Fable 5's solo score at about 63% of the cost — with Fable 5 consulted roughly once per task on average.
  • Orchestration mode: Claude Fable 5 acts as coordinator, breaking down the task and dispatching it to Claude Sonnet 5 worker sub-agents running in parallel. On BrowseComp, this combo hit about 96% of Fable 5's solo score at about 46% of the cost.
  • Both patterns rest on Anthropic's official Claude Managed Agents (multi-agent sessions) capability: advisor mode maps to a specific beta API tool (the advisor tool), orchestration mode maps to the multi-agent coordination feature — both have official docs and sample code, not something the team hacked together on its own.
  • Sub-agents and advisors each get their own independent session thread, and the thread itself persists — going back to the same sub-agent with a follow-up question doesn't mean re-feeding the whole history and paying full price again.
⚑
The original post is a first-hand account shared by Anthropic's official developer-facing account; the SWE-bench Pro / BrowseComp numbers are Anthropic's own internal benchmarks, not an independent third-party evaluation. This piece builds on the original post by digging into the corresponding official API docs and an official open-source notebook example, filling in the mechanics, code samples, and real-world pitfalls for both modes.
1Source

Who's behind this advice

Anthropic's official developer-facing account, @ClaudeDevs, shared two multi-model collaboration patterns the team uses internally on July 7, 2026.

Split the work between two Claude models at different price points: the stronger Claude Fable 5 only weighs in at key moments to advise or delegate, while the lower-priced Claude Sonnet 5 carries the task from start to finish — so most of the bill lands on the cheaper side.
⚑
The source is Anthropic's official developer-facing account, @ClaudeDevs, sharing the team's own internal practices. The original post gave only two concrete benchmark numbers (SWE-bench Pro, BrowseComp) — the mechanics need the corresponding official docs to fully explain, which the next two sections fill in.
Advisor Claude Fable 5 advice only occasional ask (thin) Executor Claude Sonnet 5 high-volume output (thick)
Signature diagram: the advisor is consulted only a handful of times (thin arrows), while the executor keeps streaming out large volumes of content (thick lines, flowing tokens).
2Mechanism One · Advisor Mode

How the advisor model and worker model split the labor

Advisor mode isn't something the team cobbled together on its own — it's a formal beta feature of the Anthropic API, the advisor tool. The executor model, Claude Sonnet 5, takes the task and works through it on its own, only consulting the advisor model, Claude Fable 5, once at a point where a judgment call is needed — then, armed with that judgment, finishes the task.

Core mechanism

Split advisor and executor roles: let the more capable Claude Fable 5 handle only "giving advice," not "writing content." All the token-heavy generation work is left to the lower-priced Claude Sonnet 5, so most of the bill lands at the executor's price point.

This isn't something you cobble together yourself by firing off two manual requests: you declare a tool of type advisor_20260301 in tools, specify which model the advisor uses, and the rest is handled entirely server-side by Anthropic. The executor decides when to ask for advice the same way it would call any ordinary tool; the server spins up a separate inference pass on the advisor model, which sees the executor's full conversation so far (system prompt, tool results, content already written), returns its judgment, and that judgment is fed straight back into the executor's conversation for it to continue writing with. The whole thing happens within a single API request: