Reasoning Model
Domain: Models & Providers
Motivation
A reasoning model emits thinking/reasoning content alongside its answer, controlled by the DeepSeek reasoningEffort levels: off, low, high, or max. low, high, and max enable thinking and serialize as output_config.effort; adapter-owned off sends thinking.type: disabled. Reasoning history is passed back verbatim into later requests.
Concrete Example
config: { reasoningEffort: high } on the DeepSeek route; an unsupported value fails with UNSUPPORTED_REASONING_EFFORT before network I/O, and session-title requests force thinking off.
Analogy
A model that shows its work before writing the final answer.