Model techniques map
Techniquesinference & servingreasoning control

specific method · filed under inference & serving

Deployment-time scalar effort control

A scalar provides flexible control of reasoning strength, with evidence describing values from 1 to 100 and preset tiers mapped onto the scale.

Also called Scalar effort control of response length, scalar efforts, Preset effort levels (low, high, max), three preset effort levels.

source
1
model
1
lab adopt it
1
strongest
default

How sources treat it

One count per evidence span, weakest treatment to strongest.

used 4default 1

Documented in

Evidence

5 spans quoted from the sources, strongest treatment first.

the public API exposes three preset reasoning-effort tiers—max, high, and low—which map directly onto this scalar interface

defaultinference servingin DeepSeek-V4.1DeepSeek

we prepend the following instruction to the system prompt: Reasoning Effort: {effort} (range 1–100; higher values request more thorough reasoning)

usedpost trainingin DeepSeek-V4.1DeepSeek

At deployment time, the scalar 𝑏 provides a flexible control interface over the model’s reasoning strength.

usedinference servingin DeepSeek-V4.1DeepSeek

the use of scalar efforts achieves flexible, interpolated control of response length within a specific range.

usedinference servingin DeepSeek-V4.1-FlashDeepSeek

we expose three preset effort levels that map onto this scale: low, high, and max correspond to effort values of 50, 75, and 100, respectively.

usedinference servingin DeepSeek-V4.1-FlashDeepSeek

Filed alongside

Other methods under inference & serving :: reasoning control.