{"slug": "run-claude-s-most-powerful-model-at-half-the-price-anthropic-s-two-official-the", "title": "Run Claude's most powerful model at half the price, Anthropic's two official patterns + the exact setup", "summary": "Anthropic published two official patterns to run its most powerful model, Claude Fable 5, at half the price by using cheaper models for routine work and reserving Fable for judgment calls. The advisor pattern scored 92% of Fable's quality at 63% of the price on SWE-bench Pro, while the orchestrator pattern achieved 96% at 46% on BrowseComp. A developer demonstrated the setup, reporting a task cost of $0.37 with only $0.14 spent on a single Fable consultation.", "body_md": "The two official patterns from Anthropic, plus the exact setup I used on camera. Benchmarks are Anthropic's own: advisor pattern scored 92% of Fable 5's quality at 63% of the price (SWE-bench Pro), orchestrator scored 96% at 46% (BrowseComp).\n\nThe idea in one line: stop letting the expensive model do the typing. Put it where judgment matters, let cheap models do the volume.\n\nYour everyday session runs on a cheap model (Sonnet). It calls Fable only when it hits something genuinely hard, about once per task.\n\nFrom your project root, create the file .claude/agents/fable-advisor.md with this content:\n\n```\n---\nname: fable-advisor\ndescription: Senior advisor for hard problems. Consult when stuck on architecture decisions, ambiguous tradeoffs, or when an approach isn't working. Returns guidance, not implementation.\nmodel: claude-fable-5\n---\n\nYou are a senior technical advisor running on Fable 5. The main session\ndoes the work. Your job is judgment, not typing.\n\nWhen consulted: analyze the problem, identify what the executor is missing,\nand return a short, decisive recommendation with reasoning. Do not write\nthe implementation yourself. Keep answers under 300 words so escalations\nstay cheap.\n```\n\nThen in Claude Code:\n\n- Restart the session (agent files load at start), or create the agent via /agents > Create new agent\n- Check /agents > Library shows fable-advisor\n- Switch your session to the cheap model: /model sonnet\n- Give it a task and tell it the advisor exists:\n\n```\nYou're the executor on this task: [your task]. When you hit an architectural\ndecision you're not confident about, consult the fable-advisor sub-agent for\nguidance, then implement its recommendation yourself.\n```\n\nReal numbers from my run: whole task $0.37, the Fable consultation was $0.14 of it, one call, exactly as the pattern describes.\n\nIf Fable's model string differs on your plan, check the exact name with /model and update the agent file's frontmatter.\n\nFlip it: Fable makes the plan and delegates the token-heavy work to cheaper Sonnet workers. The expensive model is the manager, and managers don't type.\n\nAnthropic published the exact notebook:\n[https://github.com/anthropics/claude-cookbooks/blob/main/managed_agents/CMA_plan_big_execute_small.ipynb](https://github.com/anthropics/claude-cookbooks/blob/main/managed_agents/CMA_plan_big_execute_small.ipynb)\n\nAnd the advisor tool docs:\n[https://platform.claude.com/docs/en/agents-and-tools/tool-use/advisor-tool](https://platform.claude.com/docs/en/agents-and-tools/tool-use/advisor-tool)\n\nFrom a Claude Code team member's field guide: Fable is the first model where the quality of the work is bottlenecked by how well you explain what you want. The model stopped being the limit.\n\nBefore running either pattern, run this once:\n\n```\nInterview me, one question at a time, about anything ambiguous in this task.\nPrioritize questions where my answer would change the architecture.\n```\n\nTwo more from the same guide worth stealing:\n\nBlindspot pass (before unfamiliar work):\n\n```\nI'm working on [thing] but I don't know this area well. Do a blindspot pass:\nhelp me find my unknown unknowns so I can prompt you better.\n```\n\nThe quiz (after big changes, before you merge):\n\n```\nGive me a report on everything that changed, with context and intuition,\nand a quiz at the bottom that I must pass before we merge.\n```\n\nThe model leaves subscriptions Sunday July 12. The patterns work regardless, they're how you run any expensive model without paying expensive-model prices.\n\nBuilt by @kju4q, more at github.com/kju4q", "url": "https://wpnews.pro/news/run-claude-s-most-powerful-model-at-half-the-price-anthropic-s-two-official-the", "canonical_source": "https://gist.github.com/kju4q/090ee66917a1944af4aaacc81cb6fccf", "published_at": "2026-07-08 23:36:54+00:00", "updated_at": "2026-07-09 00:10:54.930245+00:00", "lang": "en", "topics": ["artificial-intelligence", "large-language-models", "ai-products", "ai-tools", "developer-tools"], "entities": ["Anthropic", "Claude Fable 5", "Claude Sonnet", "Claude Code", "SWE-bench Pro", "BrowseComp", "kju4q"], "alternates": {"html": "https://wpnews.pro/news/run-claude-s-most-powerful-model-at-half-the-price-anthropic-s-two-official-the", "markdown": "https://wpnews.pro/news/run-claude-s-most-powerful-model-at-half-the-price-anthropic-s-two-official-the.md", "text": "https://wpnews.pro/news/run-claude-s-most-powerful-model-at-half-the-price-anthropic-s-two-official-the.txt", "jsonld": "https://wpnews.pro/news/run-claude-s-most-powerful-model-at-half-the-price-anthropic-s-two-official-the.jsonld"}}