Gemini's June leap: video, deep reasoning, and live image gen.
June 2026 was a big month for Gemini. Three things stand out: Gemini Omni pushing into video, Gemini 3 Deep Think for genuinely hard reasoning, and Nano Banana image generation now living inside Gemini Live, the real-time camera-and-voice mode. They're easy to blur together in a keynote. This sorts them out: what each is actually for, which to reach for when, and where the demo runs ahead of daily reality.
01 Three different tools, not one upgrade
The trap with a big Google drop is treating it as one giant "Gemini got better." It didn't get better in one direction — it grew in three, and each serves a different job:
| Feature | What it's for | Reach for it when… |
|---|---|---|
| Gemini Omni | Generating and working with video | You need moving images, not just stills or text |
| Gemini 3 Deep Think | Harder, slower reasoning | The problem is math, science, or multi-step logic |
| Nano Banana in Gemini Live | Real-time image creation/editing from your camera | You're pointing your phone at something and want an image now |
Keep those lanes straight and the rest of the lesson — and the tool itself — gets a lot less confusing.
02 Gemini Live + Nano Banana: image gen in the conversation
Gemini Live is the real-time mode where you talk to Gemini and share your camera. The June update folds Nano Banana — Google's image model — directly into it. Practically, that means you can generate an image from what your camera sees during a live conversation, and re-edit images Gemini already made, without leaving the flow.
Most image tools are a separate stop: open app, type prompt, wait, download. Putting generation inside a live, camera-aware conversation collapses that into one motion — point, ask, refine. That's a real workflow change for anything visual and in-the-moment: a room you're redecorating, a whiteboard, a product on a shelf.
Google shipped two new voice models on September 15, 2026: Gemini 3.8 Live (built for fast, fluid conversation with visual grounding) and Gemini 3.8 Live Extended Thinking (built to reason and speak at the same time on multi-step tasks). The Extended Thinking model is what now powers Gemini Live in the Gemini app for everyone, plus Gmail Live and Keep Live for all Google AI subscribers and Docs Live for AI Pro and Ultra. What you'll notice in practice: Live keeps talking while it runs tools in the background ("let me check that…" and then a running narration of a multi-step job), it switches between 97 languages mid-conversation on its own, and it reads what your camera sees in near real time. Search Live runs on plain 3.8 Live. Every audio reply is watermarked with SynthID.
Honest read. Google's own numbers put Extended Thinking first on Artificial Analysis' speech-to-speech quality index (82.6) and strong on voice-agent task benchmarks — but those are vendor-run evals of a model that is days old. The useful test is the one in this lesson: hand Live a multi-step task with your camera on and see whether it keeps the thread while the work runs. Business Workspace tenants get Extended Thinking "coming soon," not today.
Which task is the best fit for Nano Banana inside Gemini Live?
03 Deep Think: for when the answer needs to be earned
Gemini 3 Deep Think is the opposite of fast. It's a reasoning mode that spends more compute to work through math, science, and logic problems step by step. The trade is exactly what you'd expect: better answers on genuinely hard problems, slower and more expensive than a normal reply.
This ties into the broader Thinking Levels control Google finished rolling out — a dial for how much reasoning effort Gemini puts in. The skill isn't "always use the most thinking." It's matching the level to the task.
04 Gemini Omni: video, with the usual asterisk
Gemini Omni pushes Gemini further into video creation — generating moving footage, not just images. It's genuinely impressive, and it's also where you should keep your skeptic's hat on. AI video in 2026 is great for short, stylized, or B-roll-style clips and still wobbles on long, consistent, or factual sequences. Treat it as a fast way to draft and prototype visuals, not a finished-production button.
The honest framing for any AI video tool: it collapses the cost of a first draft. The taste, edit, and judgment about whether the clip actually works — that's still yours.
05 Which one do you actually need?
Your move
Run the same task two ways: ask a question once on a normal (fast) setting and once with Deep Think or a high thinking level. Notice both the difference in the answer and the difference in wait. That gut feel — when the extra thinking earns its cost and when it doesn't — is the whole skill with these new controls.
More Gemini lessons →What you can do now
- Keep the lanes straight: Omni = video, Deep Think = hard reasoning, Nano Banana in Live = real-time images
- Use Nano Banana in Gemini Live when the task is visual and in-the-moment with your camera
- Since September 15, 2026 Gemini Live runs on Gemini 3.8 Live Extended Thinking — expect it to keep talking while background steps run, and to switch languages on its own
- Match Thinking Levels to the task — heavy reasoning for hard, checkable problems; fast for the rest
- Treat Gemini Omni as a first-draft video tool, not finished production
- Don't pay the "more thinking" tax on simple questions a fast model already answers well