Step Star Launches Step 3.5 Flash: Low Think Mode Cuts Token Consumption by 56%

On April 2nd, Step Star announced the official launch of its latest model Step 3.5 Flash 2603, open to all Step Plan users. This update introduces a new low think mode, designed to help users utilize model resources more efficiently.

Test data shows that in the default high reasoning mode, the model maintains similar reasoning scores while reducing token consumption by 14%. When switched to low think mode, token consumption drops by 56%, making it suitable for scenarios requiring faster responses and lower reasoning complexity.