As reported by DeepMind Blog, the push for 'extended thinking' during live conversation is the notable advance here, suggesting models that can reason aloud while executing tasks. In our view, the key question is whether this capability translates outside of tightly scripted enterprise workflows to the messy, open-ended problems consumers actually face. The announcement leans heavily on benchmark scores and partner logos, but the ultimate measure will be if these models can sustain coherent, helpful dialogue when the user goes off-script or the task requires genuine world knowledge, not just API calls.