Fable 5 is Anthropic’s publicly available Mythos-class model, offering a one-million-token context window (about 750,000 words) and up to 128,000 output tokens per request. It costs $10 per million input tokens and $50 per million output tokens—twice Opus 4.8’s rate—and comes with mandatory 30-day data retention. You’ll find it on Claude.ai, Claude Code, Claude Desktop and via API under Pro, Max, Team or Enterprise plans. Three features set Fable 5 apart. First, it sustains multi-day autonomous sessions: hand it a complex project and it plans stages, delegates tasks to sub-agents and checks in only when it reaches your defined endpoint. Second, it tests its own output—writing and running checks to catch edge cases before delivering results. Third, its vision model reads dense visuals—financial charts, architecture diagrams, nested tables—with real accuracy, avoiding the misreads common in Opus 4.8. To get full value, set up a local context folder with concise markdown files: a one-page business summary, SOPs, project briefs and a running decision log. Add a self-updating memory file (claude-memory.md) where Fable records new facts and an instructions file (claude-instructions.md) that tells it how to behave every session. Point Claude Code to this folder so each session starts with your world already loaded. Finally, use the /goal and /loop commands inside Claude Code. /goal defines exactly what “done” looks like and stops Fable when it meets that target; /loop runs tasks on a schedule until you cancel it. Together they let Fable run as an autonomous worker instead of a back-and-forth chatbot. Just watch the first cycle to set cost controls—unchecked loops or goals can burn tokens fast.