@free_ai_guides - Fable 5 Setup Playbook
@free_ai_guides on X This is a practical guide to getting value from Anthropic's Fable 5 model, which the author says is wasted if you use it like a regular chat assistant. It covers a local file-based context setup, the /goal and /loop commands for autonomous work, cost-saving model routing, reusable skills, and common early mistakes.
- Fable 5 forgets everything between sessions, so the author recommends a local markdown context folder with a business summary, memory file, and instructions file that Claude Code reads at the start of each session.
- Fable 5 is priced at $10 per million input tokens and $50 per million output tokens, twice Opus 4.8, so the author suggests using it for planning, verification, and long-context work while routing routine tasks to cheaper models.
- The /goal command runs until a defined finish condition is met, and /loop runs tasks on a schedule; the author stresses specific success criteria, failure exit paths, and watching the first cycle to avoid wasted tokens.
- Requests in cybersecurity, biology, and chemistry are routed to Opus 4.8 by safety classifiers, which the author says affects under 5% of sessions, and Fable 5 requires 30-day data retention.
Fable 5 is Anthropic's Mythos-class model, built on the architecture of Claude Mythos but released publicly with safety classifiers that route high-risk queries (cybersecurity, biology, chemistry) to Opus 4.8. The author says that rerouting hits fewer than 5% of sessions, so for everyday work you're getting the full model. The specs that matter: a 1 million token context window, up to 128,000 output tokens per request, and pricing of $10 per million input tokens and $50 per million output, double Opus 4.8. It's available on the Pro, Max, Team, and Enterprise plans, and requires 30-day data retention for safety monitoring, which rules it out for work that can't be stored that long.
The author's case for Fable rests on three capabilities rather than benchmarks. First, it sustains multi-day autonomous work in Claude Code, planning across stages and delegating to sub-agents instead of needing approval at every step. Second, it writes and runs its own tests and fixes its own mistakes before reporting back, catching edge cases on the first pass that previously took two rounds with Opus. Third, its vision is noticeably sharper on charts, nested PDF tables, and technical diagrams, where Opus 4.8 would sometimes misread columns or axis labels. The article's most practical advice is the setup: Fable retains nothing between sessions, so the author recommends a local markdown context folder with a business summary, SOPs, a self-updating memory file, and a standing instructions file, all of which stay portable if you switch tools later.
The article then covers the two commands that make the model worth its price. /goal defines a finish line such as "all tests pass," and a smaller model checks after each step whether it's reached it, with the caveat that goals must be verifiable and include a failure path. /loop runs a task on a schedule until stopped, and the two can be combined. The author warns to cap costs and monitor the first cycle before walking away from a long run, since an uncapped /goal can burn tokens quickly. The rest of the piece covers cutting token costs by 60 to 80 percent through routing cheaper models to simpler work, reusable skills, and seven mistakes the author made in the first week.