ChinaStartLux’s 27B model ranks first of 173, its maker says.
A decision model routes requests, approves tool calls and attaches a confidence score for AI agents.
AutoThink answers outright at 0.9 confidence and reasons only below it, cutting a 9B model’s reasoning 65%, StartLux says.
The 73.37 versus 57.96 gap over US rival TypeSafe’s Jev comes from StartLux’s own tests.
Outside developers, who have downloaded the open weights nearly 120,000 times, can now check the claim.
How each outlet framed it
- TMTPost leans favorable
- reports StartLux's 69.1 Decision Index score and its claimed 27% lead over US firm TypeSafe's Jev, framing a Chinese startup's win as Chinese AI leadership
Sources: TMTPost