specific method · filed under inference & serving
Thinking mode selection
Thinking can be toggled on or off for a request, or models may support distinct thinking and non-thinking modes.
Also called Thinking mode, thinking, Thinking/Non-Thinking modes, dual modes (Thinking / Non-Thinking).
- sources
- 4
- models
- 4
- labs adopt it
- 3
- strongest
- default
How sources treat it
One count per evidence span, weakest treatment to strongest.
Documented in
Further reading
Picked by hand, not extracted: where to read more, not evidence for anything on this page.
- DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning paper arxiv.orgpopularized visible reasoning-trace "thinking" mode
Evidence
4 spans quoted from the sources, strongest treatment first.
Qwen3.6 models operate in thinking mode by default, generating thinking content signified by thinking ... response before producing the final responses.
M3 supports toggling thinking on or off. With thinking enabled, the model is suited to complex reasoning, agentic tasks, and long-horizon collaboration; with it disabled, it responds faster
A thinking mode can be toggled on or off per request.
Both models support 1M context & dual modes (Thinking / Non-Thinking)
Filed alongside
Other methods under inference & serving :: reasoning control.