I used to open Claude and just go with whatever model was already selected. Sometimes that worked fine. Other times I burned a pile of tokens on a task that only needed a quick answer. Or I got a thin reply on something that deserved real depth.
So I was thrilled when I found this post from an AI professional who turned the whole Claude lineup into a simple decision tree. Opus 5.5. Fable 5.1. Sonnet 5. Haiku 4.5. That’s four models and three questions, and you land on the right one every time. The author also included a ready-to-use prompt for each model, which I found really useful.
Here’s the process, step by step, with the reasoning behind each choice.
Step 1: Figure out how complex your task really is
This is the first fork in the road. Ask yourself one question: does this task require a complex answer?
- ☑️ No: use Haiku 4.5 or Sonnet 5.
- ☑️ Yes: use Fable 5.1 or Opus 5.5.
Why start here? Because this one split already saves you a lot. Most daily tasks, like rewriting an email, summarizing notes or looking up a fact, don’t need a heavyweight model. Sending them to the lighter pair keeps things fast and cheap.
Step 2: For quick tasks, choose between speed and everyday power
If you landed on the “No” side, the original poster’s next question is simple: do you need maximum speed and minimal tokens?
☑️ Yes: go with Haiku 4.5
Haiku 4.5 is the lightweight token-saver. Here are the expert’s setup tips:
- Chat without files.
- Turn on web search.
- Build in Chat. There is no more cowork anymore.
Prompt example:
I want [desired result] with [constraints]. Ask me questions using AskUserQuestion before you start.
I love this one. When Claude asks clarifying questions first, a small model doesn’t waste effort guessing what you meant. You get a better answer on the first try.
☑️ No: go with Sonnet 5
Sonnet 5 is the fast, everyday model. It’s perfect for simple tasks, and it really shines once you connect your apps with Connectors: Gmail, Google Drive, your spreadsheets, your CRM and many more.
Prompt example:
[Find/Write/Fix] this [input]. Scope: [scope]. Exactly [number] bullets, each under [length] words. Return as a [table]. No preamble.
Look at how tight that prompt is. It sets an exact bullet count, a word limit and a format, and it ends with “No preamble.” That’s how you get clean, usable output instead of a wall of filler.
Step 3: For complex tasks, ask if it’s your hardest, most ambitious work
Now for the “Yes” side. The creator puts it this way: is this your hardest, most ambitious work? Think multi-app tasks, long runs and one-shot builds.
☑️ No: go with Fable 5.1
Fable 5.1 is the deep-work model. The author’s guidance:
- Put Effort on Medium. Mostly.
- Name the finish line (and when to stop).
Prompt example:
Clean up my Downloads folder. Sort every file by type and year. Keep a checklist, work until every item is done. Don’t offer to continue: continue. Ask before deleting.
This prompt is a great lesson in naming the finish line. It defines the job and tells Claude how to track progress. It also cuts out the annoying “Want me to keep going?” pause while still requiring a check before anything gets deleted.
Once the work is done, the LinkedIn creator suggests three follow-ups:
- → Download your file.
- → Or convert it into a Claude Skill.
- → Start a fresh session to save tokens.
That last tip is easy to overlook. A fresh session means Claude isn’t hauling around a long history. That saves tokens and keeps responses snappy.
☑️ Yes: go with Opus 5.5
Opus 5.5 is the smartest model by Claude. This is where you turn on Extra effort and go for one-shot builds.
Prompt example:
Here is my goal: [goal]. Here are my constraints: [constraints]. Before acting, explore broadly: open every relevant file. Then work until every item is done.
The key is the “explore broadly” instruction. You’re telling the strongest model to gather full context before it touches anything, and that’s exactly what big builds need.
⚠️ Step 4: Be careful with Extra effort
For me, this warning from the post’s author was the most useful part. Extra effort sounds like a free upgrade, but it has real costs:
- ☑ It can run for hours (the author’s run lasted 2:40).
- ☑ Only ~20% of tasks actually need it.
- ☑ Use it for big one-shot builds, then switch to Medium.
- ☑ Long conversations slow things down. Claude sometimes re-thinks all its earlier answers.
Why it matters: more effort isn’t automatically better. If roughly four out of five tasks don’t need it, leaving Extra on by default just gets you slower answers and a bigger token bill.
Step 5: Escalate up or route down
The final rule ties everything together. When Fable gets stuck, escalate to Opus. Everything else, route down the tree.
In practice, you start with the lightest model that could reasonably handle the job and only move up when you hit a wall. A good team lead works the same way: you don’t hand a quick errand to your most senior engineer.
How to put this into practice
To make the tree stick, here’s how I’d map it to everyday work based on the author’s framework:
- Morning inbox triage: Sonnet 5 with Gmail connected, plus the bullet-limited prompt.
- Quick research question: Haiku 4.5 with web search turned on.
- Organizing a messy project folder: Fable 5.1 on Medium effort with a clear checklist.
- Building a full tool in one go: Opus 5.5 on Extra effort, then back to Medium.
The bigger lesson is that picking a model is a skill in its own right. Tokens and time cost real money, and a quick habit like this makes every Claude session faster and more focused.
Check out the full LinkedIn post for the original decision tree, and share it so your team stops burning tokens too. ♻️