THE MULTI-MODEL PICKER

Opus, Sonnet, Haiku — Claude has 3 models. Most people pick one and stick with it. Wrong move. Each is best for different kinds of work. This skill tells you which model for what — money, time, and quality optimized.

THE PROBLEM

Default behavior: people use the most expensive model for everything ("smart!"), or the cheapest model for everything ("save credits!"). Both are wrong.

The right move: match model to task. Long-context analysis = the most capable model. Quick formatting = the fastest. Most everyday work = the middle model.

THE SKILL

You give Claude:

The task you're about to do (or a category — coding, writing, analysis, quick lookup)

Claude returns:

Recommended model + reasoning
Estimated cost (relative)
Estimated time (relative)
What you'd lose by going cheaper / save by going more capable

INSTALL

Standard.

THE FULL SKILL FILE

---
name: multi-model-picker
description: Recommends which Claude model (Opus / Sonnet / Haiku — or current top/mid/fast tier) to use for a given task. Considers task complexity, time sensitivity, cost, and quality requirements. Helps users avoid both over- and under-spending on inference.
when_to_use: User asks "which model," "should I use Opus or Sonnet," or describes a task without specifying model preference.
---

# The Multi-Model Picker

You recommend Claude models based on task. Decisive. Cost-aware.

## Inputs
1. **The task** (or task category)
2. **Optional:** time sensitivity, budget, quality requirements

## Default model selection table

(Names may shift — apply the principle: top-tier / mid-tier / fast-tier)

| Task | Recommended | Why |
|---|---|---|
| Long-context analysis (Day 60) | Top-tier | Best at finding patterns in large data |
| Code generation | Top-tier or Mid-tier | Mid-tier handles 80% of tasks at lower cost |
| Daily writing (emails, posts) | Mid-tier | Quality without overspend |
| Brain dumps + organization | Mid-tier | Speed matters more than peak reasoning |
| Quick formatting / cleanup | Fast-tier | Latency matters; reasoning doesn't |
| Bulk processing (50+ items) | Fast-tier | Cost adds up at volume |
| Sales / marketing copy | Top-tier | Quality has direct ROI |
| Simple Q&A | Fast-tier | Don't burn credits for basics |
| Hard reasoning / strategy | Top-tier | Worth the spend |
| Schedule lookups | Fast-tier | Low complexity |

## Output format

🤖 RECOMMENDED MODEL: [Model name]

Why: [1 sentence]

Cost vs. alternatives:

This: ~[relative cost]
Cheapest option: ~[X% less] but you'd lose [specific capability]
Most capable option: ~[X% more] but you'd gain [specific capability]

Latency: [fast / medium / slow] Quality fit: [excellent / good / sufficient]


## Quick rules of thumb

- **If outcome dollar-value > $100:** use top-tier
- **If processing 50+ items:** use fast-tier (cost compounds)
- **If you'll edit heavily:** use mid-tier (don't pay for polish you'll redo)
- **If real-time conversation:** use fast-tier (latency matters)
- **If long context (>100k tokens):** use top-tier

## What NOT to do

- Don't recommend the most expensive model as default
- Don't recommend the cheapest for tasks where output quality has dollar value
- Don't lock the user into one model — recommend per-task switching
- Don't recommend changing models mid-conversation if context matters

## When user resists model-switching

- **"It's annoying to switch"** → Recommend setting Mid-tier as default; switch UP for deep work, DOWN for bulk only.
- **"I'll just use the best one always"** → Show the cost math. Often shocks them.

## Delivery

End with this line, exactly:

---
*Set Mid-tier as default. Switch UP for deep work, DOWN for bulk. Don't pay top-tier for formatting.*

SAFETY CHECK

Same as Day 1.

WHAT'S NEXT

Day 37 of 100. Pair with Day 62 — Token Saver (cost optimization is more than just model choice).