Claude Fable 5.1, on one card: what changed, what it costs, what to do today

Claude Fable 5.1 launched September 1, 2026. This field card is the launch in plain English: the seven things that actually changed (same $10/$50 price but cache reads cut to $0.25 so typical bills drop about 25%, and up to 45% on agentic work; long-horizon jobs; effort as the main dial including mid-conversation changes; the scoreboard vs Fable 5 and Opus 5; documents, slides and computer use; the three breaking API changes; and the relaxed cyber and biology safeguards). Then the effort cheat sheet (low / medium / high / xhigh / max, with the defaults per surface: high in Claude Code, medium in Cowork and claude.ai), the price math, the migration checklist (model id swap, remove forced tool_choice, keep history append-only, re-tune effort, re-run evals), the six prompting fixes from Anthropic's own guide, and our own test receipts: the planted 1-in-167,000 login bug hunted at high vs low effort, and the benchmark-table image read back as CSV. Independent explainer from Hyperautomation Labs; every number is sourced to the official docs.

By submitting, you'll be subscribed to the Hyperautomation AI Report newsletter — one practical AI email per week. Unsubscribe anytime.

Free. No spam. Unsubscribe anytime.