Token Optimization

Introduction

Token optimisation means saying the same thing with fewer tokens — saving cost and latency without losing quality.

Beginner analogy: Like packing a suitcase tighter. Same clothes, smaller bag.

Understanding the topic

Core concepts to understand:

• Remove redundant instructions.
• Compress few-shot examples by trimming explanations.
• Use abbreviations or codes where unambiguous.
• Trim retrieved context — rerank and keep top-K.
• Cache stable parts (system, examples) with prompt caching.

Syntax reference

Visual workflow / architecture:

Before: 1200 tokens
After:   650 tokens
Cost:    -45% / call
Quality: equal (per eval)

Real-world use

Anthropic's prompt caching can cut cost 90% on stable-prefix prompts at scale.

Best practices

• Measure tokens before & after.
• Re-run evals after trimming to verify.
• Cache aggressively.

Hands-on exercise

Interview preparation — practice these questions:

• Q1. How reduce tokens without losing quality?
• Q2. What is prompt caching?
• Q3. Trade-off of over-trimming?
• Q4. Tools for token counting?