Prompting
Temperature
A generation parameter that controls how deterministic or varied a model's output is — lower values produce more predictable text, higher values produce more diverse and surprising text.
Definition
Temperature is a scalar applied to the model's probability distribution before sampling the next token. A temperature near zero makes the highest-probability token almost always win, producing nearly deterministic output. Higher temperatures flatten the distribution, giving lower-probability tokens a real chance of being chosen. The typical range in production tools is 0 to 2, with 0.7 a common default. Temperature affects the texture of writing — its variety, surprise, and willingness to take rhetorical risks — without changing the model's underlying knowledge.
Why it matters
For writing tasks, temperature is the single most under-used control. Most users leave it at default and then complain that drafts feel samey. A temperature of 0.2 produces tight, predictable, structurally clean prose — good for factual summaries, technical writing, and anywhere consistency matters. A temperature of 0.9 produces more rhythmically varied, occasionally surprising prose — better for creative work, headlines, openings, and anywhere a draft needs a spark. The practical workflow is to match temperature to task. Lock it low for analytical work where you want reproducible output. Push it up for ideation, headline generation, and stylistic variation. Switching temperatures across phases of the same project is often more effective than searching for one optimal value.
Examples
Low temperature (0.1)
Five different runs of the same prompt produce nearly identical drafts. Good for compliance text, factual summaries, structured data extraction.
Medium temperature (0.7)
The default for most chat interfaces. Varied enough to feel alive, structured enough to stay on task.
High temperature (1.2)
Headline brainstorming or creative riffing — each run produces meaningfully different output, with some misses and some sparks.
Frequently asked
Is temperature the same as creativity?
Loosely, yes — but the model is not 'being creative,' it is sampling from a wider distribution. The variety can feel inventive or noisy depending on task and prompt quality.
What temperature should I use for editing existing text?
Low — typically 0.2 to 0.4. You want the edit to follow the source, not invent around it.
How does temperature interact with top-p?
Both control output diversity. Most practitioners adjust one and leave the other at default; combining them aggressively can produce unstable output.
Related terms
Prompt Engineering
The discipline of crafting model inputs — instructions, context, examples, and constraints — to produce reliably better outputs.
System Prompt
A persistent set of instructions that frames a model's behaviour across an entire interaction or product.
Few-Shot Prompting
Showing a model two to five worked examples of the desired input-output pattern before asking it to handle a new case.
Large Language Model (LLM)
A neural network trained on very large text corpora to predict the next token in a sequence, capable of producing fluent natural-language output.
Put the concept to work
Open the rewrite engine and apply this principle to a draft of your own.
Try a rewrite