AI glossary · Models
Temperature
In one sentence
Temperature is a language model setting that controls how much randomness goes into choosing each token. At low values the model nearly always picks the most likely option and replies predictably; at high values it varies more and comes across as more creative.
Also known as: LLM temperature · temperature parameter · sampling temperature · randomness
What is temperature?
A language model doesn’t write one fixed answer. At each step it works out a list of candidate tokens, each with a probability, and picks one. Temperature decides how that pick is made.
Put it to work
UTIO's AI, through an API
From €29/mo + VAT
For developers and agencies building their own products.
See the AI APIAt low temperature, the model almost always goes with the favourite. Replies are stable and consistent: ideal when you want accuracy. At high temperature, less likely candidates get more of a look-in. Replies vary more, with unexpected turns: handy for brainstorming, risky for stating facts.
The name comes from statistical physics, where temperature governs how much particles jiggle about. Here it governs how much the model’s choices “jiggle”.
Related to this term
AI API
- Transcription, voice, text, image, OCR and embeddings
- Per-client keys with a spending cap
- Prepaid wallet: €50 minimum top-up
From €29/mo + VAT Panel account, plus usage
See the AI APIHow it works
- The model calculates a score for every possible token.
- Those scores are divided by the temperature before being turned into probabilities.
- Below 1, the gaps widen and the favourite dominates. Above 1, they shrink and the odds even out.
- A token is picked according to those probabilities, and the process repeats.
The allowed range depends on the provider: some go from 0 to 1, others from 0 to 2. Some models, including certain reasoning models, don’t let you change it at all. As a rough guide:
| Task | Temperature |
|---|---|
| Extracting data, classifying, answering from documents | Low |
| Drafting emails or summaries | Medium |
| Brainstorming, slogans | High |
Example
A wine merchant in Bristol uses a model for two jobs. The first is pulling the wine, number of cases and delivery address out of each order that arrives by email. There it sets a low temperature: it wants the same result every time, no flair.
The second is suggesting copy for a new wine’s product page. There it turns the temperature up and asks for five versions; the team picks the best and polishes it. Same model, two settings, two uses. What it doesn’t do is raise the temperature for the tech sheet: the alcohol content and vintage are copied from the label, not generated.
Temperature at UTIO
With UTIO’s agents you don’t need to touch the temperature. What you control is what the agent knows, your knowledge base, and how much autonomy it has: no AI, ask first, draft or automatic. That’s what matters when prices or opening hours come up with customers.
If you’re building with the AI API, text generation is selected by model tier, and the full parameter reference is in the panel. Setting temperature sensibly is one of the simplest ways to cut down on made-up answers, though it’s no substitute for good sources or for checking anything important. It’s one more building block in the design of an AI agent.
Frequently asked questions
At temperature 0, does the model always give the same answer?
Almost always, but it isn't guaranteed. At 0 the model picks the most likely token, yet small numerical differences on the server can change the odd word. If you need repeatable results, fix the other settings too and store the outputs.
Does raising the temperature make the model hallucinate more?
It can. More randomness means a greater chance of picking unlikely tokens, and with them incorrect details. But lowering the temperature doesn't eliminate hallucinations: a model can make something up with complete confidence at 0.
What is top-p and how is it different from temperature?
Top-p limits the choice to the most likely tokens whose combined probability reaches a threshold, say 90%. Temperature changes how much weight each option gets; top-p trims the list of options. You'd normally adjust one or the other, not both.
Related terms
- Token
A token is the chunk of text a language model works with: a word, part of a word or a punctuation mark. Models read, count and bill in tokens, not in words or characters.
- LLM
An LLM (large language model) is an AI model trained on vast amounts of text to predict the next word. That single skill lets it draft, summarise, translate, classify and hold a conversation.
- AI hallucination
An AI hallucination is a response from an AI model that sounds convincing but is false or not backed by any source: a figure, quote, price or reference the model invents while generating the most probable text.
- Inference
Inference is when an already-trained AI model is put to use: it receives a new input, such as a question, an audio clip or an image, and computes a response from what it learned, without changing its parameters.
- Prompt
A prompt is the instruction or text you give an AI model to get something done: a question, a command, a document to summarise or a mix of all three. Much of what the model gives back depends on the prompt.
Updated on