Fable 5 is back, and it's the strongest model I've used.
It's also double the price of Opus, and it won't stay free on your Claude plan for long.
So I read Anthropic's full prompting guide, listened to their engineers, and watched what people were saying on X, then pulled it all down to six habits I'm using right now.
TL;DR
→ Fable-5 runs $10 per million input tokens and $50 per million output, double Opus
→ It's only free on your plan through July 7th, and only up to 50% of your weekly limit before you're on usage credits
→ Give it the why, tell it what not to do, and let it act once it has enough
→ Make it prove its work instead of trusting "done"
→ Stop asking it to show its reasoning, it can route you down to Opus without telling you
→ Say less, not more, because it's smart enough now that a short instruction steers just as well
The price, and the deadline
Fable 5 is not cheap. $10 per million input tokens, $50 per million output. That's double Opus.
It also won't live on your Claude plan forever. Right now it's a promotional period, and you can use up to 50% of your weekly limits with Fable-5 at no extra cost. After that you're on usage credits.
The promo ends July 7th. So you've got about six days.
It runs pretty much everywhere though. Claude desktop app, VS Code, Claude Code. Hit /model and you'll see Fable sitting there.
One thing to know before the habits: Fable follows short, clear direction better than older models because it's just better at reasoning. When you ask it to do something, it feels like it actually knows what you're talking about.
The 6 habits
1️⃣ Give it the why
Fable does better when it understands your intent. The context lets it connect your task to the right information instead of guessing what you meant.
Instead of "write me an email to a client about the delay," try "I'm working on this bigger task, here's who it's for and what they need, and with that in mind, help me write an email to this client about the delay."
If your second brain and AI OS are set up right, that context also points it at the specific files it should be reading.
2️⃣ Tell it what not to do
At its core the model is predicting the next word that makes sense, so sometimes it gets creative and does things you didn't ask for. Sometimes that's good, a lot of times it's not.
Anthropic does this all over their own guide. "When you have information to act on, act." "Don't add features, do the simplest thing that works well."
Think about briefing an intern. You tell them what not to touch because they don't know the process yet.
Instead of "take a look at this problem and handle it," try "when I'm describing a problem, the deliverable is your assessment. Report what you find and stop. Don't fix, send, edit, or delete anything until I say go."
Negative prompting used to be weak. Lately it's been working really well for me.
3️⃣ Let it act once it has enough
Stop it from over-planning. I barely use plan mode in Claude Code anymore. I build my own version by having it work until it's ready to act.
Instead of "research everything and make a full plan before you do anything," try "when you have enough information to act, then act."
This is also where effort levels matter. Low, medium, high, xhigh. Match the level to the task. Anthropic recommends high as the default, xhigh for the most capability-sensitive work, and medium or low for routine stuff.
The cost math is the part worth a look. Fable-5 on low is similar to Opus 4.8 on xhigh, and cheaper. So you probably only need to reach for Fable 5 to 15% of the time. Running it for everything is overkill, especially once you're on usage credits.
4️⃣ Make it prove it
This is probably the most important one. Models will tell you they're done when they haven't actually verified anything.
Treat it like handing work to a person. You want the verification loops baked in so deeply that when it hands you something, it's already checked its own work three or four times.
Instead of "is this done and is it working," try "before you tell me something is done, point to the result that proves it. Only report work you can show evidence for. If something isn't verified, say so plainly instead of guessing."
Bake this into your skills, your agents, and your Claude .md files instead of pasting it at the end of every prompt.
5️⃣ Stop asking it to show its reasoning
This one's Fable-specific. A standing "explain your reasoning" line, especially in a system prompt, can trigger a refusal and hand your task to Opus 4.8.
Fable ships with tighter safety guardrails. If it thinks the intent of your request might be malicious, it routes you down to a less capable model, and asking it to reveal its own reasoning can trip that wire. [Fable-specific]
6️⃣ Say less, not more
This sounds backwards because we've been trained to think more context is always better. But Fable is intelligent enough now that, wrapped in a good environment with context, tools, and skills, a short instruction steers just as well as spelling out every rule by name.
It's not a contradiction with habit one. Adding the why still doesn't mean blurting everything out.
Instead of "rule one is be concise, rule two is this, rule three is that," try "lead with the outcome, keep it simple, and pause only when the work truly needs me."
When Fable hands off to Opus
Number five points at something worth understanding on its own.
Before Fable answers, it runs a quick safety check. If your request lands in a certain bucket, hacking, dangerous biology, or asking the model to reveal its private reasoning, it pushes the task to Opus 4.8.
And it won't tell you. It routes to Opus in the background. If you're building on the API you'll see a response that shows it was Opus. Anywhere else, you might not notice.
The upside is it routes to Opus 4.8, so you're not paying Fable prices when it happens. You avoid it by not asking it to show its reasoning, and by not handing it anything that looks clearly malicious.
Wrap
Most of these aren't even Fable-specific. They're just good prompting habits that this model happens to reward more than the last one.
Bake them into your system files, your memory, and your skills once, and you stop retyping them forever.
I walk through all six with real examples in the full video. Link in the first reply.
