---
title: 'NEW Google Gemini Updates are INSANE!'
source: 'https://youtube.com/watch?v=hh6hR_RQFTM'
video_id: 'hh6hR_RQFTM'
date: 2026-11-18
duration_sec: 657
channel: 'Julian Goldie SEO'
---

# NEW Google Gemini Updates are INSANE!

> Source: [NEW Google Gemini Updates are INSANE!](https://youtube.com/watch?v=hh6hR_RQFTM)

## Summary

Google released several major Gemini updates in a single week, including the powerful Gemini 4 Carbon model, a new Decisions API for efficient automation, and the fast Gemini 4 Hydrogen model. The video provides a hands-on recap of each update, showcasing real business applications and offering practical advice on which model to use for different tasks.

### Key Points

- **Google's Ship Week Announcements** [00:02] — Sundar Pichai announced a week of major updates: new TPUs, Gemini 4 Carbon, AlphaProof code, Gemini Robotics 2.5, Alpha Genome 2, the Gemini Decisions API, Gemini 4 Hydrogen, and Intelligent Eyewear pre-orders.
- **Gemini 4 Carbon: The Strongest Model** [01:14] — Carbon is Google's most powerful model, designed for long, complex tasks like reasoning, coding, and autonomous agents. It can work on a task for hours and check its own work.
- **Carbon in Action: Booking System** [01:56] — Carbon successfully built a complete client booking and intake app for a coaching business from a single prompt, including a calendar view, intake form, confirmation, and coach dashboard.
- **Carbon with Agent Swarms** [02:22] — Using the /swarm command in Anti-Gravity, Carbon coordinated a team of agents to build a client hub and an audit report, with all agent messages visible on a shared board.
- **Carbon vs. Claude Race** [02:36] — In a five-lap race, Carbon won 3-2 against Claude. Carbon excelled at visual builds (3D landing page, dashboard, storefront), while Claude won tasks requiring high accuracy on the first try.
- **Gemini Decisions API: The Most Useful Update** [03:59] — The Decisions API doesn't generate text; it scores options and returns probabilities and confidence scores. It can handle yes/no questions, multiple-choice (up to 255 options), and rubric-based scoring, processing text, images, audio, video, and PDFs.
- **Decisions API: Practical Applications** [05:55] — The API is ideal for sorting leads, routing support tickets, scoring applications, and checking content. The confidence number allows automations to proceed when sure and escalate to humans when uncertain.
- **Setting Up the Decisions API** [06:10] — A five-step setup: get an API key, open the Decisions Playground, customize questions, test with real examples, and integrate the generated code into existing automations like Zapier or Make.
- **Gemini 4 Hydrogen: The Fast Model** [06:56] — Hydrogen is Google's fast model, 2.5x faster than Gemini 3.8 Flash, scoring 57% on Terminal Bench (vs. Flash's 18%). It's available across AI Studio, the API, the Gemini app, Google Search, and Anti-Gravity.
- **Hydrogen in Action: Orchestrator Mode** [07:20] — Using the orchestrator option, Carbon plans the work while Hydrogen executes. This was used to build client reports, a client finder kit, an onboarding hub, and a website from a brief.
- **Hydrogen Powers AI Mode in Google Search** [08:21] — Hydrogen now powers AI mode in Google Search, which fires off multiple searches simultaneously for complex questions. This has significant implications for SEO, as each search is an opportunity for pages to be cited.
- **Which Model to Use** [09:33] — Use Carbon for large, complex builds; Hydrogen for daily tasks like reports and follow-ups; and the Decisions API for sorting, scoring, and routing. Combining them creates powerful automations.

## Transcript

Gemini updates. [music] Sundar Pichai posted the list himself. Gemini 4 Carbon on Tuesday, the new Gemini Decisions API on Wednesday, [music] Gemini 4 Hydrogen on Thursday. And here's the thing, I've already built with all of it, a client
booking system, an agent swarm, client reports, websites, [music] a lead magnet quiz, and a race between Carbon and Claude. So in this video, I'll recap every update, show you what we actually built with each one, and tell you which
one to use for your business. Welcome to Ship Week. Here's the list straight from Sundar's post. What we shipped this week at Google. Monday, two new eighth-generation TPUs. Those are the chips Google trains and runs Gemini on.
chips Google trains and runs Gemini on. Tuesday, Gemini 4 Carbon plus AlphaProof code, which formally verified the signal messaging protocol and Gemini Robotics 2.5. Wednesday, Alpha Genome 2 for disease research and the Gemini
disease research and the Gemini Decisions API in AI Studio. Thursday, Gemini 4 Hydrogen and pre-orders for Google's Intelligent Eyewear. And Friday, a new U 2 album. That one was a joke. He says so himself. Three of these
you can open and use today, Carbon, Decisions, and Hydrogen. So that's where we'll spend most of our time. Let's go day by day. Tuesday, Gemini 4 Carbon. Carbon is Google's strongest model. It's built for long, hard jobs, reasoning,
coding, agents that work on one task for hours and check their own work as they hours and check their own work as they go. It's in the Gemini API and AI Studio and it's rolling out in Anti-Gravity, the Gemini app, and AI mode for Google
AI Ultra. So the first thing I did was give it a real business job. A member asked how to build their own client system instead of renting one. So I asked Carbon for a booking and client intake app for a coaching business. One
prompt, it ran on its own, and here's what came out. You your coach, pick the kind of session, and the week view shows every open slot. A clean intake form, a confirmation with a reference code, and a prep checklist. And on the coach's
side, a desk with every upcoming call, search, reschedule, and export. Then I gave Carbon a swarm. In Anti-Gravity, you can type /swarm and a whole team of
agents splits the job up and talks on a shared board. I had it build a client hub for an agency and an audit report for an AI automation consultant with every agent's message visible as it worked. And then I raced it. Gemini 4
Carbon against Claude Fable 5.2. Five laps, five real builds. Carbon 1 3. It won the visual builds. The 3D landing page, the dashboard, and the storefront
looked like finished products. But Claude won the jobs where being right the first time matters most. So my rule is simple. Use Carbon for the big visual build you'd leave running and always open the result yourself before a client
does. Now, before Wednesday's update, a quick one. If you want my exact setups from this week, the Carbon booking system prompt, the swarm briefs, the hydrogen builds you're about to see, and my Decisions API questions for leads and
client work, they're all inside the AI Profit boardroom. You also get the faceless video skill that made this video, the agent OS, and step-by-step guides for every workflow I show on this channel. Thousands of business owners
are already inside. Agency owners, coaches, course creators, SaaS founders, and every week there are new systems and live calls where you can ask me anything. The link is in the comments and the description, or go to
aiprofitboardroom.com. Wednesday, the Gemini Decisions API. This one flew under the radar, and I think it's the most useful update of the week for anyone building automations. Every AI model you've used so far writes
text. You ask a question, it writes you an answer, and then your automation has to read that answer and figure out what it meant. The Decisions API doesn't write anything. You give it a question and a set of options, and it scores
every option in one pass. You get back a probability for each one plus a confidence number between zero and one. There are three kinds of question. Yes There are three kinds of question. Yes or no? Pick one from a list up to 255
options and score against a rubric like low, medium, high. And it reads more than text, images, audio, video, and PDFs all in one request. Here's how it
looks in Google AI Studio. Open the Decisions Playground. On the home screen, you pick a question type. Ask a yes or no question, pick from fixed options, or score against a rubric. Then you add your questions. Here's a support
ticket example. Does this message describe a hardware defect? That's a yes or no. What's the right next step? That's a choice. How likely is this customer to leave? That's a score. You paste the input at the bottom, hit run,
and every question gets answered in a single call. Then you click get code and you get the code to drop straight into your own app. So what would you actually use it for? A member asked this week about agents that pull loads of data and
run up huge usage. A decision step fixes a lot of that. It sorts the work first, so only the jobs that need a big model ever reach one. If you run an agency, ask it, is this inquiry a good fit? Yes or no? And which service is it about? If
you're a coach, score every application call against your rubric. If you run a course or community, grade an assignment as pass, needs work, or excellent. And for e-commerce, show it a return photo and ask, is there visible damage? The
confidence number is the clever part. When it's sure, your automation keeps going. When it's not, it sends the job to you. Here's how to set up your one in about five minutes. Step one, open Google AI Studio and grab an API key.
There's a free tier for trying it out. Step four, open the Decisions Playground and pick one example close to your business. Step three, swap in your own questions. Keep them short and keep the options clear like good fit, maybe not a
fit. Step four, paste in five or 10 real examples from last week and check every answer and its confidence. Step five, click get code and plug it into the automation you already run, whether that's Zapier, Make, NA1, or your own
agent. Thursday, Gemini 4 Hydrogen. Hydrogen is the fast one. Google calls it the daily driver for agentic work. It finishes tasks two and a half times faster than Gemini 3.8 Flash. It scores 57% on
Terminal Bench where Flash scored 18 and on Google's own numbers, it edges past Gemini 4 Argon, the bigger model. It landed everywhere on day one. AI Studio,
the Gemini API, the Gemini app, Google Search, and Anti-Gravity. So we built a lot with it. In Anti-Gravity, I picked Gemini 4 Hydrogen from the model menu
and I used the new orchestrator option where carbon plans the work and hydrogen does the building. First, an agency client report, rankings, traffic, and next steps laid out like something you'd actually send a client. Then a client
finder kit for a coach, an ideal client profile, places to find them, and outreach messages ready to send. Then a full onboarding hub for a course owner and a first client website built from one brief. Then five more builds in one
session, a 30-day plan builder for coaches, a code base rescue where I gave it a broken lead intake agent with zero out of seven tests passing and it fixed it and a lesson turned into a lead magnet quiz. Then I put hydrogen side by
side with Argon and Flash in AI Studio on the same jobs. An agency report, a coaching follow-up, a bug in a SaaS app. You can watch every answer stream in. Hydrogen was quick and the answers held up. And I finished with a little app
that picks the right model for each job automatically. And Hydrogen is now inside Google Search. AI mode runs on it. Ask one question and it fires off a whole set of searches at once, one for each part of your question, then links
out to the sites it used. If you do SEO, every one of those searches is a chance for your page to get cited or skipped. Quickly, the rest of the week. The new TPUs are Google's eighth generation chips, one built for training and one
for running models. That's the hardware all of this runs on. AlphaProof code is a DeepMind system that writes mathematical proofs that code does what it should. They used it on the Signal protocol. Gemini Robotics 2.5 is their
model for robots, rolling out to partners first. Alpha Genome 2 predicts how changes in DNA affect specific human cells for disease research. And the intelligent eyewear is up for pre-order with Gemini built in, hands-free. It's
in stores from December 10th in the US and South Korea. So which one should you actually use? A lot of members told me this week they've started apps and abandoned them because a new model drops every two days. So here's the simple
version. Use Carbon for the big build you'd leave running, a client system, a full website, a dashboard. Use Hydrogen for everything you do every day, reports, follow-ups, quick fixes, research. Use the Decisions API for
anything that sorts, scores, or routes, leads, applications, support tickets, content checks. And if you do SEO, keep an eye on AI mode because Hydrogen now decides which pages get cited. Even better, put them together. Decisions
sorts the incoming lead. Hydrogen writes the reply. Carbon builds the system it all in. That's one automation, three models, each doing the job it's best at. My advice, pick one job in your business this week. Just one. A client reports, a
follow-up, a lead sorter. Build it with the model that fits and use it every day before you build the next thing. If you want every setup from this week, the carbon prompts, the swarm briefs, the hydrogen builds, the decisions questions
faceless video skill that made this video, it's all inside the AI Profit Boardroom. Thousands of agency owners, coaches, and course creators are building with this stuff every week and there are live calls where you can ask
me anything. The link is in the comments and the description or go to That's ship week. See you in the next one.
