Gemma-4-26B-A4B with 39-69% less thinking at same accuracy

This is the fifth model in the ThinkingCap series. We reduced how much Google's gemma-4-26B-A4B-it thinks, without compromising the quality of its answers.