Temperatur

Temperature is a parameter that controls the "creativity" or randomness of a Large Language Model's output. At temperature 0, the model always picks the most probable next token — deterministic and focused. At higher temperature (0.7–1.0), less probable tokens are also selected — more creative but less predictable.

The scale: 0 = maximally deterministic (same answer to same question every time), 0.3 = conservative (slight variations), 0.7 = balanced (default for most models), 1.0 = creative (diverse outputs), >1.0 = experimental to chaotic. Top-P (Nucleus Sampling) is a related parameter that restricts the token pool by probability.

Recommendations by use case: For facts, data, and code → temperature 0–0.3 (accuracy). For marketing copy and content → 0.5–0.7 (balance of creativity and coherence). For brainstorming and creative ideas → 0.8–1.0 (maximum variety). For image prompts → higher temperature for surprising results.

Important to know: higher temperature increases hallucination risk — the model "guesses" more boldly. For factual tasks, always use lower temperature. In most API calls, temperature can be passed as a parameter.