Temperature and Sampling Settings

What temperature does

Temperature scales randomness. Low (0–0.3) makes output focused and deterministic; high (0.8–1.2) makes it creative and varied.

Rules of thumb

• Factual, structured, code → low temperature.
• Brainstorming, creative writing → higher temperature.

Other knobs

• top_p (nucleus sampling) — an alternative to temperature; tune one, not both.
• max_tokens — cap the response length.
• stop — sequences that end generation.

client.chat.completions.create(model="gpt-4o-mini",
  messages=msgs, temperature=0.2, max_tokens=300)