Anthropic's Claude Fable 5.1 prompting guide: 15 behavior changes with fix prompts
AGI Hunt · wechat · 2026-09-02
Alongside Fable 5.1, Anthropic published an official prompting guide, noting many old prompts merely patched older models' bad habits, and listing behavior changes with copy-paste fix prompts.
Key behavior changes: fewer parallel tool calls (may degrade to one tool per turn), fewer progress updates during long calls, answering from memory under low effort, denser prose, less formatting in chat, summaries without quotes, and full-file rewrites for small edits. Each has an official fix, e.g. "privately list what you need next, then request all independent items in one response" to restore parallelism, and "you are operating autonomously" to stop mid-task questions.
Effort tiers (low/medium/high/xhigh/max, default high): start at high and tune down via your own evals; on CursorBench 3.2.0, 5.1's low matches 5's high at 1/3 the cost; cache reads drop to $0.25/MTok and effort can be switched mid-conversation without breaking cache.
Context management: treat history as append-only — editing earlier turns invalidates thinking blocks (enforced for new accounts after Aug 31); with cheaper cache reads, client-side early compression may no longer pay off. The guide also defines "mannered prose" with a removal prompt, a six-item compaction checklist (keep blockers/solutions, rejected options, decisions, progress, todos, specifics; preserve user wording verbatim), and a constraint prompt limiting unrequested code changes.
Related event: Anthropic Releases Official Prompting Guide for Claude Fable 5.1(9 posts)→
More from coding & agent
- Claude-written test harness let 31KB fake archives land on a real backup server — Grimmoner · 2026-10-18
- Open-source Tyger-S2S voice agent runs natural, interruptible conversations on a single GPU — solyarisoftware · 2026-10-18
- Matt Pocock's pre-commit hook caps agent-written files at 5,000 tokens — mattpocockuk · 2026-10-18
- Backups silently failed for 10 days: three monitoring blind spots lined up — Grimmoner · 2026-10-18
- Author Laurence Moroney: write your ugly first draft yourself, then let an LLM edit it — lmoroney · 2026-10-18
- Open-weight TYGER speech decision model makes 80ms turn-taking calls for full-duplex voice agents — solyarisoftware · 2026-10-18