WEBVTT 00:00:00.000 --> 00:00:01.480 You've probably typed something into 00:00:01.480 --> 00:00:03.560 Gemini, got an answer, and closed the 00:00:03.560 --> 00:00:05.920 tab, the same way you'd use any other 00:00:05.920 --> 00:00:08.200 chatbot. Here's the thing, that's maybe 00:00:08.200 --> 00:00:10.840 10% of what it actually does. Use it 00:00:10.840 --> 00:00:12.880 right, and it can quietly take over 00:00:12.880 --> 00:00:14.640 almost half the busy work you're still 00:00:14.640 --> 00:00:17.040 doing by hand. I spent hours mapping 00:00:17.040 --> 00:00:19.480 every model, every mode, and every 00:00:19.480 --> 00:00:22.120 product Gemini is quietly wired into, 00:00:22.120 --> 00:00:24.160 and the number that stopped me was this. 00:00:24.160 --> 00:00:27.160 Gemini's app alone has over 650 million 00:00:27.160 --> 00:00:29.320 monthly users, and that's before you 00:00:29.320 --> 00:00:31.640 count everyone using it inside search. 00:00:31.640 --> 00:00:34.200 Most of them are using maybe 20% of it, 00:00:34.200 --> 00:00:36.680 with no idea the rest even exists. Look 00:00:36.680 --> 00:00:39.040 at this data from the Census Bureau. 00:00:39.040 --> 00:00:41.400 Only about one in five US businesses 00:00:41.400 --> 00:00:44.400 actually use AI in their operations. 00:00:44.400 --> 00:00:46.040 So, if you run a business, and you're 00:00:46.040 --> 00:00:48.080 even thinking about this, you're ahead 00:00:48.080 --> 00:00:50.400 of most of your competition. 00:00:50.400 --> 00:00:51.720 What you might not know is that 00:00:51.720 --> 00:00:54.400 alongside covering AI news, we work with 00:00:54.400 --> 00:00:55.960 business owners to help them implement 00:00:55.960 --> 00:00:57.960 AI in their business. 00:00:57.960 --> 00:00:59.520 Our engineering team gets to know how 00:00:59.520 --> 00:01:01.440 your business runs, then builds the 00:01:01.440 --> 00:01:03.080 automation with you. 00:01:03.080 --> 00:01:04.440 You'll find the link in the description 00:01:04.440 --> 00:01:05.470 below. 00:01:05.470 --> 00:01:05.480 below. 00:01:05.480 --> 00:01:07.400 Click it, fill out a short form about 00:01:07.400 --> 00:01:09.240 your business, and we'll get in touch to 00:01:09.240 --> 00:01:11.440 set up a call. So, in this video, I'm 00:01:11.440 --> 00:01:14.280 breaking down exactly what Gemini is as 00:01:14.280 --> 00:01:16.240 of mid-2026, 00:01:16.240 --> 00:01:18.960 every current model, every mode, and 00:01:18.960 --> 00:01:21.000 everywhere Google has quietly built it 00:01:21.000 --> 00:01:23.280 in. By the end, you'll know exactly 00:01:23.280 --> 00:01:25.560 which Gemini tool to reach for depending 00:01:25.560 --> 00:01:27.480 on what you're actually trying to do, 00:01:27.480 --> 00:01:29.200 instead of just typing into whichever 00:01:29.200 --> 00:01:31.320 box is in front of you. First, let's 00:01:31.320 --> 00:01:33.240 clear up the biggest misconception. 00:01:33.240 --> 00:01:35.760 Gemini isn't one product at all. What 00:01:35.760 --> 00:01:38.240 Gemini actually is, here's the mental 00:01:38.240 --> 00:01:40.040 model you need before any of this makes 00:01:40.040 --> 00:01:42.880 sense. Gemini isn't a single AI, it's 00:01:42.880 --> 00:01:44.480 Google's umbrella name for a whole 00:01:44.480 --> 00:01:47.200 platform, a family of models underneath, 00:01:47.200 --> 00:01:48.880 and a set of products on top that let 00:01:48.880 --> 00:01:50.640 you actually talk to them. Think of it 00:01:50.640 --> 00:01:52.720 in two layers. The bottom layer is the 00:01:52.720 --> 00:01:54.840 models themselves, things like Gemini 00:01:54.840 --> 00:01:58.520 3.6 Flash or Gemini 3.1 Pro. These are 00:01:58.520 --> 00:02:00.920 the engines tuned for different jobs. 00:02:00.920 --> 00:02:03.000 Some built for speed, some for heavy 00:02:03.000 --> 00:02:05.760 reasoning, some for images or audio. You 00:02:05.760 --> 00:02:07.120 never see these names unless you go 00:02:07.120 --> 00:02:09.280 looking. The top layer is everything you 00:02:09.280 --> 00:02:11.959 actually click on. The Gemini app, AI 00:02:11.959 --> 00:02:14.640 mode inside Google search, Gemini inside 00:02:14.640 --> 00:02:17.040 Gmail and Docs, the voice assistant on 00:02:17.040 --> 00:02:18.680 your phone. All of those are just 00:02:18.680 --> 00:02:20.680 different doors into the same underlying 00:02:20.680 --> 00:02:22.400 models. That's the whole point of this 00:02:22.400 --> 00:02:24.680 video. Google isn't trying to build one 00:02:24.680 --> 00:02:26.560 great chatbot. It's trying to put the 00:02:26.560 --> 00:02:28.800 same AI brain behind every product you 00:02:28.800 --> 00:02:30.680 already use. So, let's start with the 00:02:30.680 --> 00:02:33.680 brains, the actual models, because once 00:02:33.680 --> 00:02:35.760 you know what each one is built for, 00:02:35.760 --> 00:02:37.800 everything else clicks into place. The 00:02:37.800 --> 00:02:39.959 current model lineup. This is a demo 00:02:39.959 --> 00:02:42.000 checklist, so we're going model by 00:02:42.000 --> 00:02:44.320 model. What it is, what it's actually 00:02:44.320 --> 00:02:46.400 good for, and where you can get it. 00:02:46.400 --> 00:02:48.840 Gemini 3.7 flash. 00:02:48.840 --> 00:02:51.600 Launched on August 13th, 2026, this is 00:02:51.600 --> 00:02:53.440 Google's newest flash model and its most 00:02:53.440 --> 00:02:55.480 capable workhorse yet. It's built 00:02:55.480 --> 00:02:58.040 primarily for coding and AI agents with 00:02:58.040 --> 00:02:59.400 major improvements in software 00:02:59.400 --> 00:03:01.519 engineering, web development, and 00:03:01.519 --> 00:03:04.320 complex multi-step workflows. Google has 00:03:04.320 --> 00:03:05.840 already made it generally available 00:03:05.840 --> 00:03:09.120 through the Gemini API, positioning 3.7 00:03:09.120 --> 00:03:11.560 flash as the new go-to model when you 00:03:11.560 --> 00:03:13.280 want strong intelligence without giving 00:03:13.280 --> 00:03:15.080 up the speed and efficiency the flash 00:03:15.080 --> 00:03:18.320 lineup is known for. Gemini 3.6 flash. 00:03:18.320 --> 00:03:20.280 This is Google's current flagship, 00:03:20.280 --> 00:03:22.320 announced in a company blog post on July 00:03:22.320 --> 00:03:25.720 21st, 2026. It's built as a workhorse, 00:03:25.720 --> 00:03:27.920 strong at coding, knowledge work, and 00:03:27.920 --> 00:03:29.920 multimodal tasks. And according to 00:03:29.920 --> 00:03:32.080 Google's own numbers, it does the job 00:03:32.080 --> 00:03:35.120 using about 17% fewer tokens on average 00:03:35.120 --> 00:03:37.320 than its predecessor. Fewer tokens means 00:03:37.320 --> 00:03:39.519 faster answers and a lower bill if 00:03:39.519 --> 00:03:41.600 you're paying for it through the API. 00:03:41.600 --> 00:03:44.080 You can reach it through the Gemini API, 00:03:44.080 --> 00:03:46.640 through AI Studio, or simply by using 00:03:46.640 --> 00:03:49.640 the Gemini app and searches AI mode. No 00:03:49.640 --> 00:03:51.519 extra setup required. If you only 00:03:51.519 --> 00:03:53.760 remember one model name from this video, 00:03:53.760 --> 00:03:55.720 make it this one because it's what most 00:03:55.720 --> 00:03:57.560 of Gemini is quietly running on right 00:03:57.560 --> 00:04:00.960 now. Gemini 3.5 Flash. This one launched 00:04:00.960 --> 00:04:03.240 back in May 2026 00:04:03.240 --> 00:04:04.800 and it's the model that was actually 00:04:04.800 --> 00:04:08.040 powering AI mode in search before 3.6 00:04:08.040 --> 00:04:10.320 Flash took over. Google described it as 00:04:10.320 --> 00:04:11.959 frontier-level intelligence at 00:04:11.959 --> 00:04:13.880 exceptional speed and by its own 00:04:13.880 --> 00:04:16.400 benchmarks, it pushed output throughput 00:04:16.400 --> 00:04:18.359 to roughly four times faster than other 00:04:18.359 --> 00:04:20.480 top models at the time. It's still very 00:04:20.480 --> 00:04:22.440 much in active use across the Gemini 00:04:22.440 --> 00:04:25.360 app, Google's anti-gravity platform, and 00:04:25.360 --> 00:04:27.560 enterprise tools. It's slightly behind 00:04:27.560 --> 00:04:30.120 3.6 now, but it's the model quietly 00:04:30.120 --> 00:04:31.919 sitting behind a huge share of what 00:04:31.919 --> 00:04:34.240 shipped this year. Gemini 3.5 00:04:34.240 --> 00:04:36.600 Flashlight. Same July announcement, 00:04:36.600 --> 00:04:38.560 different job entirely. This is the 00:04:38.560 --> 00:04:40.880 stripped-down, high-throughput sibling. 00:04:40.880 --> 00:04:43.560 Google sites roughly 350 tokens per 00:04:43.560 --> 00:04:45.919 second, which is built for volume, not 00:04:45.919 --> 00:04:47.760 depth. You wouldn't use this for a hard 00:04:47.760 --> 00:04:49.400 reasoning problem. You'd use it for 00:04:49.400 --> 00:04:51.919 background tasks and agent pipelines 00:04:51.919 --> 00:04:53.520 that need to move fast and cheap at 00:04:53.520 --> 00:04:56.520 scale. Gemini 3.1 Pro. Released in 00:04:56.520 --> 00:04:59.160 February 2026, this is the reasoning 00:04:59.160 --> 00:05:01.360 specialist. Google's benchmarks claim 00:05:01.360 --> 00:05:03.200 roughly double the logic performance of 00:05:03.200 --> 00:05:05.520 the earlier Gemini 3 Pro. It's mostly 00:05:05.520 --> 00:05:07.480 gated behind preview access through the 00:05:07.480 --> 00:05:11.360 API, anti-gravity, Vertex AI, and Pro or 00:05:11.360 --> 00:05:13.720 Ultra subscriptions in the consumer app. 00:05:13.720 --> 00:05:17.360 If 3.6 Flash is built for speed, 3.1 Pro 00:05:17.360 --> 00:05:19.520 is built for depth. The model you'd want 00:05:19.520 --> 00:05:21.760 on a genuinely hard problem, not a quick 00:05:21.760 --> 00:05:23.480 one. And if you're actually building on 00:05:23.480 --> 00:05:25.880 top of these through the API, price is 00:05:25.880 --> 00:05:27.960 where the real-world decision gets made. 00:05:27.960 --> 00:05:29.440 According to Google's own published 00:05:29.440 --> 00:05:33.000 rates, 3.6 Flash runs about $1.50 per 00:05:33.000 --> 00:05:35.600 million input tokens and $7.50 per 00:05:35.600 --> 00:05:38.240 million output tokens. For context, 00:05:38.240 --> 00:05:40.160 that's noticeably cheaper on the output 00:05:40.160 --> 00:05:44.320 side than GPT 5.6 Luna's roughly $6 per 00:05:44.320 --> 00:05:47.000 million. That gap is exactly why so many 00:05:47.000 --> 00:05:49.480 developers default to flash tier models 00:05:49.480 --> 00:05:51.960 for anything running at volume. Now, a 00:05:51.960 --> 00:05:53.680 handful of specialty models worth 00:05:53.680 --> 00:05:55.920 knowing by name, even if we don't dwell 00:05:55.920 --> 00:05:58.680 on each one. Nano Banana 2 is Gemini's 00:05:58.680 --> 00:06:00.720 current image generation and editing 00:06:00.720 --> 00:06:03.200 model, replacing the older Imagen line 00:06:03.200 --> 00:06:06.000 entirely. And that's not a small detail 00:06:06.000 --> 00:06:07.880 because Imagen is actually shutting down 00:06:07.880 --> 00:06:10.920 on August 17th, 2026. If you've had 00:06:10.920 --> 00:06:13.280 workflows built on Imagen, that clock is 00:06:13.280 --> 00:06:15.200 already running. There's also a Nano 00:06:15.200 --> 00:06:17.440 Banana 2 light variant built purely for 00:06:17.440 --> 00:06:19.919 speed, trading a small amount of quality 00:06:19.919 --> 00:06:22.000 for much faster, cheaper output at high 00:06:22.000 --> 00:06:24.640 volume. VIO 3.1 is Google's video 00:06:24.640 --> 00:06:27.120 generation model, still in beta, built 00:06:27.120 --> 00:06:29.160 to turn a text prompt into a short clip 00:06:29.160 --> 00:06:32.360 with matching audio. Gemini audio 3.5 00:06:32.360 --> 00:06:34.360 live translate handles real-time 00:06:34.360 --> 00:06:36.919 speech-to-speech translation across more 00:06:36.919 --> 00:06:39.360 than 70 languages, already built into 00:06:39.360 --> 00:06:41.160 Google Meet and Android. And if you're 00:06:41.160 --> 00:06:42.919 curious about the more niche end of the 00:06:42.919 --> 00:06:45.080 lineup, there's a security-focused 00:06:45.080 --> 00:06:48.200 variant called 3.5 flash cyber built to 00:06:48.200 --> 00:06:50.000 coordinate with vulnerability scanning 00:06:50.000 --> 00:06:53.520 tools. And Lyra 3.5, Google's music 00:06:53.520 --> 00:06:55.720 model, which can now generate tracks up 00:06:55.720 --> 00:06:58.000 to 3 minutes long from a text prompt. 00:06:58.000 --> 00:06:59.600 Here's the honest limitation worth 00:06:59.600 --> 00:07:01.560 naming. Google ships a lot of these 00:07:01.560 --> 00:07:03.520 models fast, and the naming gets 00:07:03.520 --> 00:07:07.560 confusing on purpose or not. 3.5, 3.6, 00:07:07.560 --> 00:07:11.160 3.1 pro, flashlight, flash cyber. If 00:07:11.160 --> 00:07:12.960 you're not building on top of the API 00:07:12.960 --> 00:07:15.320 professionally, you genuinely don't need 00:07:15.320 --> 00:07:17.480 to memorize this list. You just need to 00:07:17.480 --> 00:07:19.720 know the shape of it. Fast and cheap, 00:07:19.720 --> 00:07:22.000 deep reasoning, and multimodal. That's 00:07:22.000 --> 00:07:23.880 really three categories wearing a lot of 00:07:23.880 --> 00:07:26.040 different name tags. The modes you 00:07:26.040 --> 00:07:28.400 actually interact with, models are the 00:07:28.400 --> 00:07:30.600 engine. Modes are the steering wheel. 00:07:30.600 --> 00:07:32.080 Here's where things get useful for 00:07:32.080 --> 00:07:34.760 anyone who isn't a developer. AI mode 00:07:34.760 --> 00:07:36.600 inside Google Search turns your search 00:07:36.600 --> 00:07:40.360 bar into a conversation. As of IO 2026, 00:07:40.360 --> 00:07:42.960 it's globally powered by Gemini 3.5 00:07:42.960 --> 00:07:45.560 flash and instead of 10 blue links, you 00:07:45.560 --> 00:07:47.040 get a written answer with follow-up 00:07:47.040 --> 00:07:49.240 questions and sometimes an interactive 00:07:49.240 --> 00:07:51.160 widget built on the fly. Anyone with 00:07:51.160 --> 00:07:53.000 search can use it. No subscription 00:07:53.000 --> 00:07:55.240 required. Ask something like, "What's a 00:07:55.240 --> 00:07:57.120 quick dinner with what's in my fridge?" 00:07:57.120 --> 00:07:59.320 and it answers in full sentences, not a 00:07:59.320 --> 00:08:01.720 list of recipe blogs. Deep Think is the 00:08:01.720 --> 00:08:04.080 extra effort version of the Gemini app. 00:08:04.080 --> 00:08:06.120 It spends more compute per answer to 00:08:06.120 --> 00:08:07.840 reason through harder problems 00:08:07.840 --> 00:08:10.080 step-by-step. Google gates this one 00:08:10.080 --> 00:08:12.280 behind Google AI Ultra and it's built 00:08:12.280 --> 00:08:14.160 for genuinely difficult science or 00:08:14.160 --> 00:08:16.200 engineering questions, not everyday 00:08:16.200 --> 00:08:18.760 chat. Now, Deep Research is where this 00:08:18.760 --> 00:08:20.840 stops being a chatbot and starts being 00:08:20.840 --> 00:08:22.960 an assistant. You give it a topic and 00:08:22.960 --> 00:08:24.960 instead of one reply, it plans a 00:08:24.960 --> 00:08:27.680 research strategy, opens web pages, 00:08:27.680 --> 00:08:30.600 reads them and if you allow it, pulls 00:08:30.600 --> 00:08:32.800 from your own Gmail and Drive, too. What 00:08:32.800 --> 00:08:34.840 comes back isn't a paragraph. It's a 00:08:34.840 --> 00:08:37.320 full multi-page report inside Gemini's 00:08:37.320 --> 00:08:39.760 canvas. This is the part of Gemini that 00:08:39.760 --> 00:08:42.479 actually earns the word agent and we're 00:08:42.479 --> 00:08:44.320 coming back to why that matters in a few 00:08:44.320 --> 00:08:46.640 minutes. Gemini Live is the voice and 00:08:46.640 --> 00:08:49.120 camera mode. Say, "Hey Google, let's 00:08:49.120 --> 00:08:50.560 chat" and you're talking to it 00:08:50.560 --> 00:08:52.640 hands-free with the option to point your 00:08:52.640 --> 00:08:54.360 camera at something and ask what it's 00:08:54.360 --> 00:08:56.720 looking at live. And Canvas is the 00:08:56.720 --> 00:08:59.320 workspace mode. Type, "Create a quiz app 00:08:59.320 --> 00:09:01.720 about planets" and it writes the code, 00:09:01.720 --> 00:09:03.720 the interface and the content in one 00:09:03.720 --> 00:09:06.240 pass, right there for you to edit. One 00:09:06.240 --> 00:09:08.320 more worth a mention briefly because 00:09:08.320 --> 00:09:10.680 it's still early. Gemini Spark, a 00:09:10.680 --> 00:09:14.080 personal agent announced at IO 2026, 00:09:14.080 --> 00:09:15.520 meant to run continuously in the 00:09:15.520 --> 00:09:16.960 background handling things like 00:09:16.960 --> 00:09:19.000 scheduling. Right now, it's limited to 00:09:19.000 --> 00:09:21.280 early Ultra testers. So, treat this one 00:09:21.280 --> 00:09:23.680 as coming, not here. Quick gut check 00:09:23.680 --> 00:09:25.880 before we move on. If all of that sounds 00:09:25.880 --> 00:09:27.720 like a lot of separate tools, that's 00:09:27.720 --> 00:09:29.880 fair. But, notice the pattern. Every 00:09:29.880 --> 00:09:31.960 single one of these modes is just Gemini 00:09:31.960 --> 00:09:35.160 3.5 or 3.6 flash wearing a different job 00:09:35.160 --> 00:09:37.240 title. You're not learning six different 00:09:37.240 --> 00:09:39.560 AIs, you're learning six different ways 00:09:39.560 --> 00:09:41.400 to ask the same brain for help. 00:09:41.400 --> 00:09:43.720 Multimodal in practice. Let's talk about 00:09:43.720 --> 00:09:45.400 what multimodal actually means 00:09:45.400 --> 00:09:47.000 day-to-day, because it's more than a 00:09:47.000 --> 00:09:49.360 buzzword on a slide. Gemini reads and 00:09:49.360 --> 00:09:51.720 writes text and code. Obviously, that's 00:09:51.720 --> 00:09:53.680 the baseline. But drop a photo into a 00:09:53.680 --> 00:09:56.120 chat and ask it to caption or edit it, 00:09:56.120 --> 00:09:58.240 and Nano Banana handles that. Ask it to 00:09:58.240 --> 00:10:00.480 speak an answer out loud, and Gemini's 00:10:00.480 --> 00:10:02.400 audio models generate that voice on the 00:10:02.400 --> 00:10:05.040 spot with actual control over tone and 00:10:05.040 --> 00:10:07.680 pacing. Ask for a short video and VO 00:10:07.680 --> 00:10:09.880 builds one from scratch. Ask it to edit 00:10:09.880 --> 00:10:12.520 an existing clip, swap the sky, change 00:10:12.520 --> 00:10:14.480 the style, and that's a separate tool 00:10:14.480 --> 00:10:17.080 called Gemini Omni doing frame-by-frame 00:10:17.080 --> 00:10:19.360 editing by voice command. Inside Google 00:10:19.360 --> 00:10:21.760 Docs and Sheets, the same underlying 00:10:21.760 --> 00:10:23.560 models can draft a document from your 00:10:23.560 --> 00:10:25.880 meeting notes or build a spreadsheet out 00:10:25.880 --> 00:10:28.160 of a pile of invoices, complete with 00:10:28.160 --> 00:10:30.400 formulas and charts. Not just raw 00:10:30.400 --> 00:10:33.160 numbers dumped into cells. In Slides, 00:10:33.160 --> 00:10:35.040 hand it a list of bullet points and it 00:10:35.040 --> 00:10:37.280 can lay out an actual deck, not just 00:10:37.280 --> 00:10:39.560 text on blank slides. And through Gemini 00:10:39.560 --> 00:10:41.560 Live's camera mode, you can point your 00:10:41.560 --> 00:10:43.520 phone at a menu in a language you don't 00:10:43.520 --> 00:10:45.400 speak and get a live translation 00:10:45.400 --> 00:10:47.280 overlaid on what you're looking at. Or 00:10:47.280 --> 00:10:49.080 ask it to identify an object it's 00:10:49.080 --> 00:10:51.120 looking at through the lens. No typing 00:10:51.120 --> 00:10:52.720 involved. Here's the part worth 00:10:52.720 --> 00:10:55.040 remembering. You never pick the model. 00:10:55.040 --> 00:10:56.920 You just say what you want. Make this an 00:10:56.920 --> 00:10:59.720 infographic. Translate this. Write this 00:10:59.720 --> 00:11:02.200 in Python. And Gemini quietly roots the 00:11:02.200 --> 00:11:04.280 request to whichever model actually does 00:11:04.280 --> 00:11:06.480 that job. That's the design philosophy 00:11:06.480 --> 00:11:09.080 in one sentence. One platform, and it 00:11:09.080 --> 00:11:10.880 decides the plumbing so you don't have 00:11:10.880 --> 00:11:13.480 to. Where Gemini actually lives. This is 00:11:13.480 --> 00:11:15.560 the part that's easy to underestimate. 00:11:15.560 --> 00:11:18.120 Gemini isn't confined to one app. It's 00:11:18.120 --> 00:11:19.839 spread across nearly everything Google 00:11:19.839 --> 00:11:22.560 ships. In Search, it's AI mode, already 00:11:22.560 --> 00:11:24.960 covered. In Gmail, it's behind Smart 00:11:24.960 --> 00:11:27.440 Compose and auto-reply suggestions. In 00:11:27.440 --> 00:11:30.000 Docs, Sheets and Slides, Ultra and Pro 00:11:30.000 --> 00:11:30.710 Pro 00:11:30.710 --> 00:11:30.720 Pro 00:11:30.720 --> 00:11:33.000 get Gemini drafting text, building 00:11:33.000 --> 00:11:35.840 formulas, and designing slide layouts, 00:11:35.840 --> 00:11:37.800 pulling context from your own files when 00:11:37.800 --> 00:11:40.120 you let it. In Drive, it can find and 00:11:40.120 --> 00:11:42.280 summarize documents for you. In Google 00:11:42.280 --> 00:11:43.760 Meet, it's doing live caption 00:11:43.760 --> 00:11:46.120 translation. On Android, especially 00:11:46.120 --> 00:11:48.360 Pixel devices, it's baked straight into 00:11:48.360 --> 00:11:49.920 the voice assistant. And there's a 00:11:49.920 --> 00:11:51.680 Chrome extension that lets the browser 00:11:51.680 --> 00:11:54.080 send page content straight to Gemini, so 00:11:54.080 --> 00:11:56.000 you can ask questions about whatever tab 00:11:56.000 --> 00:11:58.280 you're on. And for developers, all of it 00:11:58.280 --> 00:12:00.680 is exposed through Google AI Studio and 00:12:00.680 --> 00:12:03.280 the Gemini API, plus a newer platform 00:12:03.280 --> 00:12:05.120 called antigravity for building 00:12:05.120 --> 00:12:07.320 multi-agent workflows on top of it. The 00:12:07.320 --> 00:12:09.360 strategic point here isn't subtle. 00:12:09.360 --> 00:12:10.920 Google isn't trying to win the best 00:12:10.920 --> 00:12:13.080 standalone chatbot argument. It's trying 00:12:13.080 --> 00:12:14.600 to make sure you're never more than one 00:12:14.600 --> 00:12:16.880 product away from Gemini, no matter what 00:12:16.880 --> 00:12:18.720 you're doing on a Google device or in a 00:12:18.720 --> 00:12:21.160 Google app. Agents, the part that 00:12:21.160 --> 00:12:23.280 actually matters. Now, here's the shift 00:12:23.280 --> 00:12:24.960 I promised earlier, the one that 00:12:24.960 --> 00:12:26.880 actually changes what this platform is 00:12:26.880 --> 00:12:29.600 for. Everything so far has been ask a 00:12:29.600 --> 00:12:31.960 question, get an answer. Agents are 00:12:31.960 --> 00:12:33.880 Google trying to move Gemini past that 00:12:33.880 --> 00:12:36.280 entirely. Deep research is the clearest 00:12:36.280 --> 00:12:39.280 example already live, plan, browse, 00:12:39.280 --> 00:12:41.360 synthesize, write, without you 00:12:41.360 --> 00:12:43.720 babysitting every step. Spark is the 00:12:43.720 --> 00:12:45.480 early, still limited attempt at a 00:12:45.480 --> 00:12:47.360 persistent personal agent running 00:12:47.360 --> 00:12:49.480 continuously in the background. And on 00:12:49.480 --> 00:12:51.560 the developer side, antigravity lets 00:12:51.560 --> 00:12:53.440 companies build coordinated teams of 00:12:53.440 --> 00:12:55.839 sub-agents. Google's own blog post gave 00:12:55.839 --> 00:12:57.360 an example of businesses running 00:12:57.360 --> 00:12:59.360 parallel agents to analyze data at 00:12:59.360 --> 00:13:01.600 scale, rather than one model doing 00:13:01.600 --> 00:13:03.440 everything sequentially. Picture the 00:13:03.440 --> 00:13:05.960 difference in practice. The old way, you 00:13:05.960 --> 00:13:08.120 ask Gemini, "What should I know before a 00:13:08.120 --> 00:13:09.800 trip to Japan?" and it gives you a 00:13:09.800 --> 00:13:12.600 paragraph. The agent way, you say, "Plan 00:13:12.600 --> 00:13:14.560 my trip to Japan." and it checks 00:13:14.560 --> 00:13:17.080 flights, compares hotel options, and 00:13:17.080 --> 00:13:19.240 comes back with an actual itinerary, 00:13:19.240 --> 00:13:20.880 pausing to confirm with you before it 00:13:20.880 --> 00:13:22.560 books anything. That's the same 00:13:22.560 --> 00:13:24.640 underlying model, just given permission 00:13:24.640 --> 00:13:26.400 to take more than one step before 00:13:26.400 --> 00:13:28.240 handing control back to you. None of 00:13:28.240 --> 00:13:30.120 this is science fiction anymore, and 00:13:30.120 --> 00:13:31.960 none of it is fully finished, either. 00:13:31.960 --> 00:13:34.000 That's the honest read. Deep Research 00:13:34.000 --> 00:13:36.440 genuinely works today. Spark is still in 00:13:36.440 --> 00:13:38.520 early testing, but the direction is 00:13:38.520 --> 00:13:40.800 unmistakable. Google wants Gemini to 00:13:40.800 --> 00:13:42.960 eventually take a task, break it into 00:13:42.960 --> 00:13:45.520 steps, and execute most of them without 00:13:45.520 --> 00:13:47.400 you typing a follow-up for every single 00:13:47.400 --> 00:13:49.440 one. What actually makes Gemini 00:13:49.440 --> 00:13:51.280 different? So, how does this stack up 00:13:51.280 --> 00:13:53.000 against everyone else building the same 00:13:53.000 --> 00:13:55.560 kind of thing? Let's be balanced here, 00:13:55.560 --> 00:13:57.520 because Google's advantages are real, 00:13:57.520 --> 00:13:59.720 but so are its weak spots. The clearest 00:13:59.720 --> 00:14:02.120 edge is data. Gemini can pull from live 00:14:02.120 --> 00:14:05.440 search results, Maps, Gmail, and Drive 00:14:05.440 --> 00:14:08.200 in ways that a closed sandbox chatbot 00:14:08.200 --> 00:14:09.960 simply can't match without plugins 00:14:09.960 --> 00:14:12.440 bolted on. The second edge is reach. 00:14:12.440 --> 00:14:14.280 Every Android phone is a potential 00:14:14.280 --> 00:14:16.480 Gemini client, and every Workspace 00:14:16.480 --> 00:14:18.080 business account already has it 00:14:18.080 --> 00:14:20.520 available. No competitor has that kind 00:14:20.520 --> 00:14:22.480 of built-in distribution. And on raw 00:14:22.480 --> 00:14:25.240 benchmarks, Gemini 3 Pro topped the LM 00:14:25.240 --> 00:14:27.680 Arena leaderboard, which, regardless of 00:14:27.680 --> 00:14:29.360 how much weight you put on any single 00:14:29.360 --> 00:14:30.839 leaderboard, says Google's 00:14:30.839 --> 00:14:33.000 infrastructure and DeepMind's research 00:14:33.000 --> 00:14:35.480 are producing real, top-tier results, 00:14:35.480 --> 00:14:38.000 not just hype. But, and this matters for 00:14:38.000 --> 00:14:40.320 credibility, Google is genuinely more 00:14:40.320 --> 00:14:41.920 conservative about rollout than some 00:14:41.920 --> 00:14:44.040 competitors. Deep Think and Spark are 00:14:44.040 --> 00:14:46.000 still gated behind ultra subscriptions 00:14:46.000 --> 00:14:48.040 or limited testing, while some rivals 00:14:48.040 --> 00:14:49.680 ship new capabilities to everyone at 00:14:49.680 --> 00:14:52.200 once. And the tier structure itself, 00:14:52.200 --> 00:14:56.560 free, pro, ultra, API pricing, can be 00:14:56.560 --> 00:14:58.480 genuinely confusing next to a simpler 00:14:58.480 --> 00:15:00.600 flat subscription from a competitor. If 00:15:00.600 --> 00:15:02.120 you've ever opened the Gemini pricing 00:15:02.120 --> 00:15:04.240 page and closed it 5 minutes later still 00:15:04.240 --> 00:15:06.240 unsure which plan you need, that's not 00:15:06.240 --> 00:15:08.520 just you. Where this is actually headed, 00:15:08.520 --> 00:15:10.280 a few things are confirmed and a few are 00:15:10.280 --> 00:15:12.000 still rumor, and it's worth keeping 00:15:12.000 --> 00:15:14.920 those separate. Confirmed, Gemini 3.5 00:15:14.920 --> 00:15:16.760 Pro is currently in partner testing with 00:15:16.760 --> 00:15:18.839 a public release expected soon, and 00:15:18.839 --> 00:15:20.240 Google has already started training on 00:15:20.240 --> 00:15:23.440 Gemini 4, according to its own July 2026 00:15:23.440 --> 00:15:25.000 announcement, though there's no public 00:15:25.000 --> 00:15:27.160 timeline for that yet. Workspace AI 00:15:27.160 --> 00:15:29.600 rollout continues expanding, and Gemini 00:15:29.600 --> 00:15:31.280 Live's regional language support keeps 00:15:31.280 --> 00:15:33.920 growing. Speculative and worth labeling 00:15:33.920 --> 00:15:35.720 clearly as such, there's talk of a 00:15:35.720 --> 00:15:38.080 dedicated on-device AI chip for future 00:15:38.080 --> 00:15:40.040 Pixel phones, and some experimental 00:15:40.040 --> 00:15:42.480 DeepMind research around 3D avatars and 00:15:42.480 --> 00:15:44.600 world simulation that hasn't shipped as 00:15:44.600 --> 00:15:46.440 a product. Treat both of those as 00:15:46.440 --> 00:15:49.280 possible, not coming. Nothing official 00:15:49.280 --> 00:15:51.600 has confirmed either one. The verdict. 00:15:51.600 --> 00:15:53.640 So, where does that leave things? Gemini 00:15:53.640 --> 00:15:56.160 in 2026 isn't a chatbot you occasionally 00:15:56.160 --> 00:15:58.440 open. It's an AI layer Google has 00:15:58.440 --> 00:16:00.840 threaded through search, Gmail, your 00:16:00.840 --> 00:16:03.400 documents, and increasingly your phone 00:16:03.400 --> 00:16:05.880 itself. The models handle the thinking, 00:16:05.880 --> 00:16:08.200 the modes handle how you ask, and agents 00:16:08.200 --> 00:16:10.040 like Deep Research are the clearest sign 00:16:10.040 --> 00:16:11.960 of where all of it is actually heading. 00:16:11.960 --> 00:16:13.400 If there's one thing worth trying this 00:16:13.400 --> 00:16:15.520 week, it's Deep Research on something 00:16:15.520 --> 00:16:17.040 you'd normally spend an evening looking 00:16:17.040 --> 00:16:19.160 into yourself, and actually watching it 00:16:19.160 --> 00:16:20.760 work instead of just reading the final 00:16:20.760 --> 00:16:22.720 report. Drop a comment with which piece 00:16:22.720 --> 00:16:24.640 of this surprised you most, the model 00:16:24.640 --> 00:16:27.320 lineup, the agent side, or just how much 00:16:27.320 --> 00:16:28.760 of this you were already using without 00:16:28.760 --> 00:16:30.480 realizing it. I'll be back soon with a 00:16:30.480 --> 00:16:32.400 deeper breakdown on how Deep Research 00:16:32.400 --> 00:16:34.000 actually performs against a real 00:16:34.000 --> 00:16:36.080 research task. Thanks for watching, and 00:16:36.080 --> 00:16:39.200 I'll see you in the next one.