Fable 5.1：完整指南（作者：Anthropic）

1. 业务流程 (AutomationBench)：17.1% → 31.4% —— 几乎翻倍。
2. 智能体科学研究：24.7% → 52.6%
3. 智能体编码 (Terminal-Bench 4.0)：42.0% → 55.8%
4. 计算机使用：36.1% → 41.7%

• 缓存读取：每百万 token 从 $1.00 降至 $0.25（降价 75%）
• 典型工作负载：整体便宜约 25%
• 重度智能体工作负载：最高便宜 45%
• 输入定价：每百万 token $10 —— 与 Fable 5 相同。
• 输出定价：每百万 token $50 —— 与 Fable 5 相同。

• 在长时程、困难任务上表现更佳。
• 缓存读取成本降低75%。输入和输出定价未变，但长时间运行的代理会话（跨多轮保持上下文）现在成本显著降低。
• 它旨在解决根本原因而非快速修复。
• 总体上，Claude Code 中的误报（以及网络安全中的假阳性）更少。

1. 管理努力级别

2. 进度更新提示

PROGRESS UPDATE PROMPT  "Before you start, say in a line what you're about to do; brief  updates while you work help the user follow along. Close with a  short recap that stands on its own - what you found, what you  did, and what's next."  Also worth checking: if you've got an old instruction telling it  to "hold findings for the final response," remove it. That was a  workaround for older models being too chatty. It now suppresses  updates you probably want.

3. 写作密度

"I prefer concise outputs - keep responses direct and avoid
unnecessary elaboration."

• Fable 5.1使用的项目符号和格式样式要少得多。如果你更喜欢这种旧的Fable 5风格，只需说明即可。
• 赋予模型自主权。从总体方向开始，然后让Fable 5.1自行运行（这正是该模型的全部意义所在）。
• 在xhigh和max effort模式下，为长输出留出空间

/claude-api migrate this project to claude-fable-5-1

MIGRATION PROMPT

"I want to migrate this project from [Fable 5 / Opus 5 / Opus 4.8]
to Claude Fable 5.1.

Scan the codebase and identify every place the model is called,
then walk me through what needs to change:

- Swap the model ID to claude-fable-5-1
- Flag any use of forced tool use (tool_choice set to a specific
tool or 'any') so I can update it to tool_choice: auto
- Check whether my code edits conversation history in any way -
injecting reminders into old messages, trimming turns, changing
the system prompt mid-session - since Fable 5.1 rejects or drops
thinking blocks when that happens
- Check my data retention settings support the 30-day requirement
Fable 5.1 needs

Show me the full list of changes before making any edits, and tell
me which ones you can apply automatically versus which ones need
my manual review."

SKILL UPGRADE PROMPT

"I have a Claude skill I built for an earlier model. I want to
check whether it needs updating for Fable 5.1.

Read through the skill file and flag anything that might now be
wrong or outdated, specifically:

- Any instruction telling the model to hold updates until the end,
or suppress progress narration - Fable 5.1 already writes fewer
updates by default, so this makes it worse
- Any anti-formatting rules like 'avoid bullet points' or 'don't
use headers' - Fable 5.1 under-formats by default now, so old
suppression rules may need loosening or removing
- Any instruction that assumes the model checks in before acting -
Fable 5.1 is built to finish tasks autonomously, so permission-
seeking language may need rewriting into scope and approval
boundaries instead
- Any assumption about how verbose or dense the writing should be,
given Fable 5.1's prose runs denser by default
- Any place the skill relies on forced tool use, since that no
longer works on this model

For each issue you find, tell me what it currently does, why it's
likely wrong for Fable 5.1, and suggest the exact replacement text.
Don't change anything yet - just show me the list first."

1. 思考究竟是什么

2. 推理提示

3. 系统提示

4. 重试