Temperature
A model configuration parameter that controls how random or deterministic an AI model's outputs are. Low temperature produces consistent, predictable responses. High temperature produces more varied, creative outputs. For most operational automation, low temperature is the right default.
What is Temperature?
Temperature is a numerical setting — typically between 0 and 2 — that controls how the AI model samples from its probability distribution when choosing the next token. At temperature 0, the model always picks the most probable next token: outputs are deterministic and consistent across runs. At temperature 1, the model samples proportionally from its probability distribution, introducing variation. At temperature 2, low-probability tokens are sampled more aggressively, producing highly varied and sometimes incoherent outputs.
Mathematically, temperature scales the logits (raw prediction scores) before the softmax function converts them to probabilities. Higher temperature flattens the distribution, giving less likely options a greater chance. Lower temperature sharpens it, concentrating probability on the most likely choice.
When to Use Low vs. High Temperature
Temperature 0 (or near 0): Data extraction, document classification, structured output generation, ERP field population — any task where consistency and accuracy matter more than variety
Temperature 0.3–0.7: Summarization, question answering, drafting business communications — tasks where some natural variation is acceptable but factual accuracy still matters
Temperature 0.8–1.2: Creative copy generation, brainstorming, generating multiple variants of ad text or email subject lines
Temperature above 1.2: Rarely useful in production. Outputs become unpredictable and often degraded.
Temperature in Operations
For operations teams, the default answer is almost always low temperature. An AI agent extracting invoice line items should return the same result every time it sees the same document — temperature 0. An agent classifying inbound supplier emails as urgent, standard, or informational should not vary its classification based on random sampling. Temperature is one of the first configuration questions to answer when deploying an AI workflow: how much variation can this task tolerate? For most operational automation, the answer is none.