This website uses cookies

Read our Privacy policy and Terms of use for more information.

Welcome back to the AI Edge (by Miles Deutscher) weekly newsletter!

Anthropic just released Fable 5.2, and it's a big upgrade.

It's faster, 20% cheaper than Fable 5.1, and significantly better at coding, research, and complex tasks.

Anthropic also released its official prompting playbook, explaining how to get the most out of the new model.

This week, we're breaking down the most important changes, with prompts you can copy to improve your results.

After our main guide, stick around for the Midweek Edge for this week's AI news and Looking Ahead, where we compile our top AI research, trends, and workflows.

WEEKLY BUILD & WORKFLOW

Fable 5.2 TL;DR

TL;DR: Fable 5.2 is faster, cheaper, and considerably more capable across technical and non-coding tasks, according to Anthropic's benchmarks.

Here are some of the biggest improvements over Fable 5.1:

→ Real-world freelance tasks: 17.9% → 41.3%

→ Robotics: 27.6% → 64.9%

→ Business workflows: 31.4% → 51.4%

→ Agentic coding: 32.4% → 49.1%

→ Chart reading: 44.8% → 73.4%

Pricing:

Fable 5.2 is also 20% cheaper than Fable 5.1.

→ Input: $10 → $8 per million tokens

→ Output: $50 → $40 per million tokens

→ Cache reads: $0.20 per million tokens

→ Fast Mode: Up to 2.5x faster, at 16/80 per million input/output tokens.

It also supports a 1M-token context window and up to 128K output tokens.

Choosing Effort Levels

Most of Fable 5.2's capability is already available at High. Going above it doesn't always produce better results.

Here's the recommended setup (by Anthropic):

→ Low: Quick questions and simple tasks.

→ Medium: Everyday work and interactive coding.

→ High: Difficult coding, research, and agent work.

→ Xhigh: Your hardest long-running coding and agent tasks (multi-hour tasks).

→ Max: Open-ended science, mathematics, and research.

High is the default through the API, while Claude apps use Medium by default.

Prompts to Improve Fable 5.2

Your existing Fable 5.1 prompts will mostly work, but Anthropic recommends a few adjustments.

Here are three worth saving:

1. Finishing long tasks

On longer tasks, Fable 5.2 sometimes stops early, lists what's left, or offers to continue instead of finishing.

So, add this ‘finisher’ prompt to your instructions:

Don't stop while there's still work I asked for. 

Keep going on anything that doesn't depend on my answer.

If you hit ordinary obstacles (errors, timeouts, empty results, failing tools), diagnose them and work through them: retry, fix the request, or try another tool. 

Don't stop just to offer to continue.

If something is a deliberate blocker (a file marked off-limits, access I've withheld, or a safety guardrail), leave it alone, tell me what you found, and look for another safe way to finish. 

Still ask before any risky or irreversible action.

2. Progress Updates

Fable 5.2 often goes “quiet” during long tasks.

You can get progess updates using this prompt:

Before you start, tell me in one line what you're about to do.

Give brief updates as you work. 

Finish with a short recap that makes sense on its own: what you found, what you did, and what's next.

3. Final Reports

Anthropic recommends this “final report” prompt:

Give me a final report of [task].

Put the key points first. Use plain words and full sentences, and don't use terms you made up during the task.

Use bullets only for lists of separate items.

Keep it concise, but include the context needed to understand it.

Migrating to Fable 5.2

Prompt to migrate any workflows for Opus 5.5/Fable 5.1 to Fable 5.2:

I want to migrate this project from [Fable 5.1 / Opus 5.5 / other]
to Claude Fable 5.2.

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

- Swap the model ID to claude-fable-5-2

- Flag anywhere my code edits conversation history (changing the
  system prompt or tools mid-conversation, trimming or rewriting
  earlier messages), since Fable 5.2 rejects invalid history

- Flag any router or fallback that moves conversations between
  models, since only Fable 5.2 and Mythos 5.2 can read Fable 5.2's
  thinking

- Flag any use of the computer_20251124 tool so I can move to the
  new computer use toolset

- Flag any advisor tool setup so the advisor model can be set to
  claude-fable-5-2

- Flag any forced tool use (tool_choice set to "any" or a specific
  tool) so I can switch to "auto"

Show me the full list of changes before editing anything, and tell
me which ones you can apply automatically versus which need my
review.

Note: If you only use Claude through the regular app, you don't need to handle this migration manually.

Upgrade Your Claude Skills

Some Skills written for older models may now be costing you more than necessary.

Use this Skill Upgrade Prompt to review your existing setup:

I have a Claude skill built for an earlier model. Check whether it
needs updating for Fable 5.2.

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

- Instructions telling the model to think less or answer without
  deliberating. These no longer work at Medium effort and above.
  I should lower effort instead

- Standing instructions to delegate to subagents, which can
  drive up cost

- Token budgets written into the prompt, which can make it spend
  more

- Hard style rules (like strict sentence-length limits), which it
  now follows so closely that writing can get choppy

- Absolute "never" rules or cautious notes that could make it hold
  back routine work

- Vague approval rules that should name exact actions

For each issue, tell me what it does now, why it's likely wrong for
Fable 5.2, and suggest replacement text.

Don't change anything yet. Show me the list first.

You can then review the recommended changes before updating your Skills.

Final Tips

A few things to remember before you start:

→ Don't use Fable for everything. We recommend using Fable as an orchestration model for smaller subagents like Haiku/Opus.

→ Keep your prompts simple.

→ Review old Skills. The big one. Small changes to existing instructions can improve Fable's behavior and reduce unnecessary token spend.

MIDWEEK EDGE: Manually curated list of the most important weekly AI news updates

Weekly AI News Recap

→ OpenAI Launches GPT-6 Cyber. A new cybersecurity model.

→ ChatGPT Images 2.6 Is Live. OpenAI's latest image model generates images 50% faster, with improved editing and support for animated GIFs.

→ Google Releases Gemini Omni 1.2 Flash. The updated video model supports clips up to 15 seconds, follow-up editing, and lower pricing.

→ Meta Launches Muse Spaces. You can now create custom dashboards, trackers, and interactive tools inside Muse using natural-language prompts.

→ Claude Can Now Connect to Bank Accounts. Anthropic's new Personal Finance feature lets you analyze spending and transactions from connected accounts. Rolling out in beta to eligible US subscribers.

→ Meta Launches Muse Ember. Meta's smallest AI model yet.

Available now through the Meta Model API.

LOOKING AHEAD: Manually curated list of useful AI tools, skills, workflows & more

Tools and Research Worth Your Time

→ 10 Free GitHub Repositories to Upgrade Claude.

A collection of open-source tools for Skills, memory, agent management, and development workflows.

→ AI Task Length over the years (METR Time Horizon)

In 2025, the best models could handle tasks that would take a human around 6.5 hours. When Mythos was released in 2026, that number 2.5x'd to 16+ hours.

→ AI Singularity is here

22 AI leaders claiming the singularity is here (or said we were close months ago)

→ A summary of Claude 2.0 (new features, upgraded subagents & more):

CONTENT PIPELINE: Latest content from AI Edge

New From AI Edge

YouTube: 6 Ways to Print $$ with Grok + Claude. Watch now →

𝕏: Claude 2.0: Everything That's Changed (FULL GUIDE). Read now →

Skool: Free prompts, cheat sheets, PDF guides & more. Join here (100% free) →

See you next Wednesday,
The AI Edge Team

Want to sponsor or partner with us? Reach out at [email protected].