Temperature, in AI, means a setting that controls how random an AI’s answers are. Low temperature gives safer, more predictable wording; high temperature gives more varied and surprising wording, with more risk of going off track.
Think of a friend recommending a restaurant. On a cautious day they always name the same reliable favourite. On an adventurous day they might suggest the brand-new place down the road. Temperature is how adventurous the AI is allowed to be when it picks each next word. It has nothing to do with heat.
In a sentence: “For the product descriptions I turned the temperature down, so the AI kept the wording plain and consistent.”
What the word means
The plain point: an AI writes by picking one likely next word after another. Temperature decides whether it nearly always takes the most likely word, or sometimes takes a less likely one.
Anthropic’s glossary describes it this way: “Temperature is a parameter that controls the randomness of a model’s predictions during text generation.”1
Google’s glossary says temperature changes “the randomness of token selection”, which affects how creative or predictable the answer is.2 A token is a small piece of text, often part of a word. See token meaning.
A simple example
The plain point: the same question can come back plain or playful, depending on the setting.
| Setting | Asked for a bakery name |
|---|---|
| Low temperature | Safe: it suggests “The Corner Bakery”. Ask again and you get something very similar. |
| High temperature | Varied: it might suggest “Crumb & Moon”, then something quite different the next time. More ideas, and more misses. |
Low is for consistency. High is for variety. Neither one makes the AI more truthful.
Why it matters to you
The plain point: most people never touch it. Chat apps like ChatGPT, Claude and Gemini set it for you. You only meet it in developer tools and settings panels.
- Zero does not mean identical every time. Anthropic warns that even at the lowest setting, the same question can give different answers.1
- Some vendors now say: leave it alone. Google deprecated the setting in its Gemini API on 21 July 2026. “Deprecated” means the company has announced it is ending support.3 The full story is in Gemini temperature setting deprecated.
- Anthropic has gone further on its newest models. Its developer docs say the setting is “not supported on Claude 4.7 and later models and Claude Mythos Preview”: a request that sets it to anything but the default is refused with an error.4
- Words often work better than the dial. Want variety? Ask for “four options, each with a different structure”. Want consistency? Give an example of the output you want. That is this site’s practice, as explained on the Gemini page above.
Often confused with
- Accuracy: turning temperature down does not stop the AI making things up. It only makes its wording more predictable. See hallucination meaning and prompting myths, which checks the “turn it down for facts” advice.
- Top_p and top_k: two related settings that also limit which words the AI may pick. Google deprecated them at the same time as temperature, and Anthropic’s docs list them alongside temperature as not supported on Claude 4.7 and later models and Claude Mythos Preview.4
- Reasoning effort: a different setting on newer models. It controls how much the AI works through a problem before answering, not how random its wording is. See reasoning model meaning.
Questions people ask
Does temperature 0 give the same answer every time?
Not always. Anthropic says that even at 0, “identical inputs may produce different outputs across API calls”.1
Does a higher temperature make AI more creative?
It makes the wording more varied. Google’s glossary says temperature influences “the creativity and predictability of the output”. More varied is not the same as more accurate.2
Can you still change the temperature?
Less and less. Google deprecated the setting in its Gemini API on 21 July 2026,3 and Anthropic says it is not supported on Claude 4.7 and later models.4
Learn more on this site
- The Gemini change: Gemini temperature setting deprecated: what changed.
- Common mix-ups: AI terms people get wrong.
- Testing a change properly: how to test whether a prompt change actually helped.
- The basics: the AI glossary and AI explained for complete beginners.
QUICK CHECK · 3 QUESTIONS
Pick an answer to see if you got it. Nothing is recorded.
1. What does the temperature setting control?
- How random the AI’s wording is
- How hot the computer gets while it runs
- How many facts the AI knows
Show the answer
Answer: How random the AI’s wording is. The page says temperature decides whether the AI nearly always picks the most likely next word or sometimes a less likely one.
2. Does turning the temperature down make the AI more truthful?
- Yes, a low setting stops made-up facts
- No, it only makes the wording more predictable
- Yes, but only for short answers
Show the answer
Answer: No, it only makes the wording more predictable. The page says low is for consistency and high is for variety, and neither makes the AI more truthful.
3. Want more variety without touching a dial? What does the page suggest?
- Restart the app until the answer changes
- Type the whole question in capital letters
- Ask for several options with different structures
Show the answer
Answer: Ask for several options with different structures. The page says words often work better than the dial: ask for options with different structures, or give an example for consistency.
Related terms: AI literacy · AI memory · Sentiment analysis · Sycophancy · System prompt · Zero-shot prompting
- 1Anthropic, Glossary, undated documentation, read at source 1 Oct 2026“Temperature is a parameter that controls the randomness of a model’s predictions during text generation.” ↑a“Even with temperature set to 0, the results will not be fully deterministic and identical inputs may produce different outputs across API calls.” ↑b“Even with temperature set to 0, the results will not be fully deterministic and identical inputs may produce different outputs across API calls.” ↑c
- 2Google Cloud, Generative AI glossary, last updated 24 Sep 2026, read at source 1 Oct 2026“Temperature changes the randomness of token selection during response generation, which influences the creativity and predictability of the output.” ↑a“Temperature changes the randomness of token selection during response generation, which influences the creativity and predictability of the output.” ↑b
- 3Google, Gemini API release notes, entry dated 21 July 2026, read in a browser 16 Sep 2026, re-read directly at source 13 Oct 2026“Deprecated parameters: The sampling parameters temperature, top_p and top_k are now deprecated.” ↑a“Deprecated parameters: The sampling parameters temperature, top_p and top_k are now deprecated.” ↑b
- 4Anthropic, Using the Messages API, undated documentation, read at source 5 Oct 2026“The
temperature,top_p, andtop_ksampling parameters are not supported on Claude 4.7 and later models and Claude Mythos Preview. Setting them to a non-default value returns a 400 error.” ↑a“Thetemperature,top_p, andtop_ksampling parameters are not supported on Claude 4.7 and later models and Claude Mythos Preview.” ↑b“Thetemperature,top_p, andtop_ksampling parameters are not supported on Claude 4.7 and later models and Claude Mythos Preview.” ↑c
Google, Gemini API release notes, entry dated 21 July 2026, read in a browser 16 Sep 2026 (as cited on this site’s Gemini temperature page); re-read directly at source 13 Oct 2026.
The comparisons on this page (the cautious or adventurous friend, the bakery names) are this site’s own way of explaining, not claims from these sources.
Free printable AI cheat sheets, and AI news every week.
Sixteen one-page AI guides and checklists in one free PDF, no sign-up. And if you want to keep up, the newsletter covers what changed in AI each week, in plain English, with the sources.
SUBSCRIBE FREE ↗ GET THE CHEAT SHEET PACK