User asks how to limit reasoning effort on Qwen 3.5 and Gemma 4

A user is seeking methods to control the reasoning chain length for Qwen 3.5 and Gemma 4 models within an agentic coding platform.

The goal is to prioritize speed at low settings while ensuring the model fully utilizes its capacity at high or xhigh settings. The user notes that controlling the 12B Gemma 4 model via system prompt is relatively easy, but finds it awkward with other models.