Stage 1 of 9 · 6 lessons · ~1 week

Start from the task, not the model

There is no best model, only a best model for a job. Describe the job first: what kind of task it is, how good the answer must be, how fast, how often and with what data. Then use leaderboards to build a shortlist, not to make the decision.

Start: Five kinds of task

Lessons· one lesson for each topic, in this order

Do this

Name the task family of your feature
Write the six requirement lines: quality, speed, volume, context, tools, privacy
Write down what a wrong answer costs you
Read one blind-vote leaderboard and one benchmark site
Pick three candidate models of different tiers or providers
Date your shortlist and link the pages you read

Start with· follow one of these

CoursePaid
Grokking Modern AI Fundamentals
Design Gurus · Beginner · $72 one-time
A structured grounding in how models, tokens and embeddings work before you compare them
GuideFree
Applied LLMs: What We Learned From a Year of Building
Yan, Bartolo, Fu and others · Advanced
Free: practitioners on starting with a strong model, then measuring before you step down

Practise

Google AI Studio
Try the same prompt on several Gemini tiers in the browser and feel the trade-off yourself

Keep handy

Arena (formerly LMArena)
Blind-vote rankings by task type: use it to build a shortlist
Artificial Analysis
Measured intelligence, speed, latency and price per model and per provider

Also worth your time

Choosing the right model (opens in a new tab)Read
A provider's own short guide to matching capability, speed and cost to the task
AI EngineeringGo deeperPaid
The chapters on evaluation and model selection are the most thorough treatment in print

Before moving on: A one-page requirements card for a feature you want to build: its task family, the six requirement lines, and a shortlist of three candidate models taken from two leaderboards, with the date you read them.

Tauseef's advice

Most bad model choices happen before any testing: someone picks the model they read about last week. Ten minutes writing down quality, speed, volume, context, tools and privacy saves weeks. Then let leaderboards narrow the field, and let your own data decide. For a shorter walk through the same decision, read sefism.com/questions/how-to-choose-an-llm-for-my-project/.