Temperature, Top-p, and Controlling Randomness

What is Temperature?

Temperature controls how creative vs predictable Claude's responses are. It's a number between 0 and 1.

Temperature = 0

Most predictable. Always picks the highest-probability word. Great for factual extraction, classification, code generation.

Temperature = 0.5

Balanced. Some variety while staying coherent. Good default for most production use cases.

Temperature = 1.0

Most creative. More surprising word choices. Use for brainstorming, creative writing, generating diverse options.

Think of it this way

Temperature 0 = Claude always takes the highway (fastest, most direct). Temperature 1 = Claude explores side streets (interesting, sometimes gets lost).

What is Top-p (Nucleus Sampling)?

Top-p controls the pool of words Claude considers. Instead of looking at ALL possible next words, it only considers the top words whose combined probability reaches p.

Setting	Value	Effect	Use Case
Top-p = 0.1	Very narrow	Only top ~10% of words	Precise, factual output
Top-p = 0.5	Moderate	Top ~50% of words	Balanced variety
Top-p = 0.9	Wide	Almost all words	Creative, diverse output

Pro tip

Don't change both temperature AND top-p at the same time. Anthropic recommends adjusting temperature and leaving top-p at default. They interact in complex ways.

Practical Settings Guide

Task	Temperature	Top-p	Why
Data extraction	0	1 (default)	Need exact, consistent output
Code generation	0 - 0.2	1	Correctness over creativity
Customer support bot	0.3	1	Consistent but natural
Marketing copy	0.7	1	Creative but coherent
Brainstorming	1.0	1	Maximum variety

# Python example - setting temperature
import anthropic
client = anthropic.Anthropic()

# Factual extraction - temperature 0
response = client.messages.create(
    model="claude-sonnet-4-20250514",
    max_tokens=1024,
    temperature=0,
    messages=[{"role": "user", "content": "Extract the date from: 'Meeting scheduled for March 15th'"}]
)

# Creative writing - temperature 0.8
response = client.messages.create(
    model="claude-sonnet-4-20250514",
    max_tokens=1024,
    temperature=0.8,
    messages=[{"role": "user", "content": "Write a tagline for a coffee shop"}]
)

Key Takeaways

This lesson covered temperature, top-p, and controlling randomness
Apply these concepts in your own projects before moving on
Refer back to this lesson when you encounter related challenges

← Back to Course Next Lesson →