# A few patterns we frequently use with Fable 5 (thread) > [!question] Lingering Questions > - What is the absolute (not relative) cost of the all-Fable-5 baseline these percentages are measured against? > - How was "~once per task" advisor invocation enforced or measured? > - Do the patterns compose — an orchestrator whose workers themselves use an advisor? > [!tip] Article Ideas > - Advisor-vs-orchestrator decision snippet for agent builders: which pattern fits which task shape. > [!link] Cross Reference > - [[3. Resources/Permanent Notes/Route Judgment to the Expensive Model and Bulk Execution to the Cheap One]] — this note already captures both patterns from this thread; see Permanent candidates below rather than a new note > - [[3. Resources/Permanent Notes/Tool-Interface Choice Shapes Agent Architecture More Than Model Choice]] — the routing decision above the model is the real lever > [!quote] Vocabulary > - *Advisor pattern* — cheap executor calls the expensive model rarely, for guidance only > - *Orchestrator pattern* — expensive model plans and delegates volume to cheap worker sub-agents ## Metadata - Source: [[3. Resources/Readwise/Full Document Contents/Tweets/A few patterns we frequently use with Fable 5]] - Author: @ClaudeDevs - URL: https://twitter.com/ClaudeDevs/status/2074606058128224365#thread ## Author's thesis Two named cost-routing patterns — advisor and orchestrator — let a multi-model agent system keep most of Fable 5's quality while paying mostly Sonnet 5's rate, because the expensive model is metered to rare judgment calls rather than high-volume execution tokens. ## Key arguments 1. **Advisor pattern**: an executor (Sonnet 5) does the bulk of the work and calls Fable 5 only occasionally for guidance — ~92% of Fable 5's SWE-bench Pro score at ~63% of the price. 2. **Orchestrator pattern**: Fable 5 plans and delegates to Sonnet 5 workers — 96% of Fable 5's BrowseComp performance at 46% of the price. 3. **Per-sub-agent caching** is the mechanism that keeps repeated delegation affordable — each sub-agent keeps its own cache, so repeat calls don't re-pay for the same context. ## Claim atoms ### Advisor pattern bills most tokens at executor rate > [!quote] ID:rw_01kx4nkgbehd6mvphy3d6sj5q0 > Use Fable 5 as an "advisor." An executor (Sonnet 5) calls Fable 5 for guidance. Most tokens are billed at the lower executor rate. Where/why: opening pattern in the thread; core cost mechanism behind rare expensive-model calls. #xref ### Advisor holds ~92% quality at ~63% cost > [!quote] ID:rw_01kx4nkgbehd6mvphy3d6sj5q0 > Sonnet 5 + a Fable 5 advisor tool gets ~92% of Fable 5's score at ~63% of the price. Where/why: SWE-bench Pro result; single-benchmark, vendor-reported. #xref ### Orchestrator holds 96% quality at 46% cost > [!quote] ID:rw_01kx4nkgbehd6mvphy3d6sj5q0 > The Fable 5 orchestrator achieves 96% of Fable 5 performance at 46% of the price. Where/why: BrowseComp result with Sonnet 5 worker sub-agents; better cost/quality point than advisor here, but a different benchmark. #xref ### Per-sub-agent cache avoids double-paying for context > [!quote] ID:rw_01kx4nkgbehd6mvphy3d6sj5q0 > Each sub-agent keeps its own cache, so repeat calls don't pay in full for the same context twice. Where/why: closing detail on Claude Managed Agents; explains why repeated delegation stays cheap. #xref ## Permanent candidates - **Already captured**: [[3. Resources/Permanent Notes/Route Judgment to the Expensive Model and Bulk Execution to the Cheap One]] states this thread's core claim (route judgment to the expensive model, volume to the cheap one) and cites both the advisor and orchestrator quotes directly. No new Permanent created from this source — connected instead. ## Processing checklist - [ ] Find the absolute cost baseline behind the "% of price" figures - [ ] Compare advisor vs. orchestrator on the *same* benchmark to see which frontier actually wins - [x] Connect into the relevant MOC — [[3. Resources/MOCs/AI Cost & Tokenomics MOC]] ## Links - [[3. Resources/MOCs/AI Cost & Tokenomics MOC]] — direct cost/quality tradeoff data for multi-model agent routing - [[3. Resources/Permanent Notes/Route Judgment to the Expensive Model and Bulk Execution to the Cheap One]] — evidence-for; this note's central claim