Technical Context
I looked at Google's latest announcements, and my impression immediately split in two: Omni makes you want to get your hands on it, while Gemini 3.5 Flash already looks like a solid base for AI automation. These are different updates in terms of readiness, and they shouldn't be confused.
The Omni story is still raw. From what has surfaced in products and demos, it's a new video direction on the level of seedance 2: generation, remixing, templates, possibly more coherent transitions, and better scene retention. But I don't see an API, nor any public, stable documentation. This means for my AI architecture, it's not a building block yet, but an interesting signal of where Google is taking multimodality.
Gemini 3.5 Flash, on the other hand, looks grounded and useful. It's pitched as a fast model, and if the 280+ speed figure holds up under real load, it's interesting not just on a slide but in production. Plus, there's an important point: it outperforms the previous Pro version on some benchmarks. Not everywhere, not magically, but the direction is clear.
This is where I paused. When a fast-tier model is immediately available via API, it's far more important than a fancy demo. I can quickly check its latency, tool calling, how stably it handles long chains, and how it behaves in data extraction, request routing, and agentic scenarios.
The Antigravity CLI is amusing in its own right. It seems Google is building a new layer of dev tools around its models and workflows. If they can make the CLI convenient, old habits around the Gemini CLI could genuinely shift towards this new entry point.
What This Changes for Business and Automation
In short, Gemini 3.5 Flash is the winner here. I'd look at it for support, internal assistants, classification, summarization, multimodal intake, and cheap agentic chains where speed and cost matter more than record-breaking reasoning. This already looks like a proper artificial intelligence implementation, not just a showcase feature.
For now, Omni only wins in one area: it raises the bar of expectations for video generation within the Google ecosystem. But without an API, it's not the tool I would base a client pipeline or AI solution development on with a clear SLA.
The losers here are those who build plans based on rumors. I've seen it many times: a cool video inspires, and then it turns out there's nothing to integrate. That's why at Nahornyi AI Lab, we usually first build a working circuit with available models and only then add new ones when they truly become part of the stack.
If you're hitting a wall with response speed, inference cost, or need to carefully embed AI integration into your current processes, let's break it down for your stack. At Nahornyi AI Lab, I ground these things without magic: I select the model, build the architecture, and help you build AI automation that lives in production, not just in a demo.