Anthropic finally published the prompting playbook for every Claude model. > Prompting Claude Sonnet 5. > Prompting Claude Opus 4.8. > Prompting Claude Opus 5. > Prompting Claude Fable 5. After reading it, one thing is clear. Almost nobody is getting the most out of Claude. It shows exactly... how each model should be prompted instead of relying on guesswork. If Claude is in your workflow, bookmark this.show more

the real gains come from understanding how each one thinks and structuring the workflow around it

You can finally test each Claude model with its intended prompting style instead of one generic harness. Hold the task constant, then compare success rate, retries, latency, and cost.

GEM

I wanna to read is right now

Reading prompt guides is one thing. The real gap is that most teams still treat prompts as strings instead of small programs with versioning, evals, and diffs. That's the practice most people are skipping.

The per-model split is the useful part. Most prompting guides age badly because they assume one model behaves like the next — this one doesn't.

appreciate the share, finally a clear guide to squeeze the most out of each Claude model

good share

why claude make it auto ???

Interesting, prompt engineering is still quite underrated I would say. yet people are jumping on graphs lol

感觉很多人都是把 Claude 当普通聊天框用

Still, no one will read it. But keep complaining about the results.

essential reading for anyone using opus

Official prompting guides are usually worth reading. The biggest improvement often comes from understanding what each model is designed to do, rather than trying to force the same prompt to work everywhere

anyone using Claude should read this if they want better results

now I know more about Claude

Now everything has become crystal clear

Understanding the model beats guessing

model specific prompts matter

this playbook fixes model steerability instantly

thanks bro

A good reminder that model capability is only part of the equation. Understanding how to prompt a model for its intended strengths can make a surprisingly large difference in real-world results.

the Fable 5 section is interesting. curious how it compares to Opus in practice

Framing is what most people skip—jumping straight to the task without context is why outputs feel generic. Try testing different prompt strategies across models on

bookmarked cant wait to step up my prompting game with Claude

Definitely a game-changer for Claude users.

Wow, this is huge! Thanks for sharing. I'm definitely going to dive into this to see how I can improve my prompting.

they cut over 80 percent of the old system prompts with no loss on coding evals

Great, I’ll let ChatGPT know about this.

Now anyone can replicate this

> Prompting ChatGTP

fable 5 in playbooks shows prompting adapts to new model versions

Four model-specific prompting docs. 80% of Anthropic's own code is already written by Claude. They shipped the guide after they stopped needing it

what's the most surprising tip in there?

I will consider it when start my dev session tomorrow