AI Pulse · Frontier Labs Watch

Google has released Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, splitting its offer into two versions aimed at different jobs. One is positioned for real-time, multimodal work. The other is for tasks that need deeper reasoning before the model responds. Google is clearly drawing a line between speed-sensitive interactions and more considered analysis, as outlined in the company’s release.

For a business owner, the important point is not the model name. It’s that AI buying is becoming more like buying cloud infrastructure. You may not want to pay for deep reasoning when a customer service assistant needs to respond quickly. Equally, you probably don’t want a fast, lightweight response for work such as analysing a complex proposal, reviewing a long document, or working through a planning problem.

Google’s two-tier approach positions it against OpenAI’s own real-time and reasoning options. That gives enterprise buyers another reason to stop treating AI as one generic subscription. The practical question is which workflows need immediate interaction, which need accuracy and careful analysis, and what each category costs to run at scale.

This is the kind of thing we build into an AI command centre, routing work to the right capability rather than asking one model and one price tier to handle everything.