StudyAIStudyAI
Pro
NLP BasicsLesson 6
Lesson 68 min

Modern NLP: Small Models vs LLMs

Choosing the right tool for cost, speed, and complexity.

What you will learn
  • When a small model wins
  • When to use an LLM
  • Balancing cost, speed, and accuracy

Explanation

LLMs are powerful but slow and costly. Many NLP tasks don't need them.

For high-volume, well-defined tasks (spam, sentiment, routing), a small fine-tuned model is faster and far cheaper. For complex, open-ended tasks (summarising nuanced text, answering varied questions), an LLM shines.

A smart system often routes: a cheap model handles the easy 90%, and an LLM handles the hard cases.

Real-world use

A support tool might classify tickets with a tiny model and only call an LLM to draft replies for tricky ones — cutting cost dramatically.

Common mistakes
  • Defaulting to an expensive LLM for every task, including ones a small model handles cheaply.
Practice

List two NLP tasks best done by a small model and two best done by an LLM, with reasons.

Knowledge check
0/2 answered

1. For a high-volume, simple task you should usually prefer...

2. When does an LLM make most sense?

Answer all questions to check.