Temperature (Sampling)
TL;DR
Controls how creative or predictable AI responses are.
What does this mean?
Temperature controls how “random” text generation is. Low temperature (0) = deterministic and focused. High temperature (1+) = more creative and variable.
How it works
For each generated token, the model selects from a probability distribution. Temperature 0 always takes the most probable word; higher values increase the chance of surprising alternatives.
Example
For contract analysis: low temperature (precise). For creative marketing copy: higher temperature (more variation).
Why it matters
Getting the temperature setting right is critical — too high leads to nonsense, too low leads to rigid repetition.
Related terms