Physix Frontier · News Briefing Card (TMTPost · Oct 25, 2026)

Kimi launches K3.1 Pro, terminal task score jumps from 12.6% to 67.2%

KEY FACTS

Kimi released K3.1 Pro, with the standard version notably faster than K3 and a new Highspeed edition added.
K3.1 Pro's Artificial Analysis composite index rose from 44 to 55.
On the Terminal-Bench 4.0 terminal task test, the score climbed from 12.6% to 67.2%, ranking first.
API output pricing: K3 at 100 yuan per million tokens, Pro at 68 yuan, and Highspeed at 136 yuan.
In testing, the model was able to correct clipping and walking-path errors in 3D scenes on its own.

KEY DATA

44 to 55Composite index gain

12.6% to 67.2%Terminal task score

68 yuan per million tokensPro API output price

136 yuan per million tokensHighspeed output price

PHYSIX OBSERVATION

K3.1 Pro's advances on terminal tasks and physical simulation show Kimi is extending from a language model toward a foundation model for embodied intelligence. The API price cut and the Highspeed edition are meant to attract developers, but opaque quota consumption could hurt the experience. If it keeps improving physical understanding, Kimi may be able to build a differentiated edge in the AI Agent race.

Source: TMTPost report