Data
Temperature (AI Parameter)
Temperature is a numerical parameter (typically 0–2) that controls the randomness of an AI model's output. At temperature 0, the model deterministically selects the highest-probability token at each step, producing consistent, conservative outputs. At higher temperatures, lower-probability tokens are sampled more often, producing more varied and creative — but potentially less accurate — outputs.
For enterprise AI deployments, temperature is a tunable reliability dial. Customer-support and document-extraction tasks benefit from low temperature (0–0.3) for consistency. Creative writing and brainstorming tasks use higher temperature (0.7–1.0) for variety. Production systems handling regulated tasks (clinical notes, financial analysis, legal drafting) should default to low temperature and test any increase against accuracy benchmarks before deploying.
Temperature (AI Parameter) in practice
Temperature is one of several settings tuned during testing, alongside the prompt, the model and output length limits. A practical method is to fix the evaluation set first, compare a few temperature values on it and keep the lowest one that still gives acceptable variety. Some reasoning models restrict or ignore the setting, so check the provider's documentation. These settings are part of the evaluation work in LLM API integration.
Related terms
- LLM (Large Language Model) · Data
- Inference · Data
- Hallucination · Data
- Fine-Tuning · Data