Generation parameters and what replaced them

temperature, top_p and top_k are gone on current models. Knowing what to use instead is the exam-relevant part.

5 min read · Lesson 9 of 12 in this domain

Older Claude models exposed temperature, top_p and top_k to control sampling randomness, and a great deal of published material still assumes they are available. On current models they are removed and return a 400. This matters twice over: as a migration fact, and because the question behind the parameter still needs answering. If you reached for temperature: 0 for determinism, the replacement is a tighter prompt and a lower effort setting — noting that temperature zero never guaranteed identical output anyway. If you reached for high temperature to get variety, the replacement is prompting for it explicitly, for example asking the model to propose several distinct directions and then pick one.

• temperature, top_p and top_k return a 400 on current models. Remove them rather than tuning them.
• On Claude 4.x models that still accepted them, you could send either temperature or top_p, never both.
• For determinism: tighten the prompt and lower effort. Temperature zero was never a guarantee of identical output.
• For variety: ask for it in the prompt — several distinct options, then a choice — rather than relying on a sampling knob.
• effort lives inside output_config and is the current control for reasoning depth and overall token spend.
• Fixed budget_tokens thinking is likewise removed on current models; adaptive thinking replaces it.

Source: Claude Docs — Model migration guide · Independent study aid, not affiliated with or endorsed by Anthropic.