GPT-6 Astra, Fable 5.1, and Gemini 3.8 Flash: The Week Three AI Labs All Shipped
OpenAI, Anthropic, and Google all released flagship models in the same week, Gemini Notebook quietly capped how much you can generate, and Grok's new payments plugin lets a bot spend your money with your sign-off.
The most capable model rarely becomes the one people actually use every day; this week OpenAI, Anthropic, and Google all shipped flagship releases, but daily adoption is decided by cost and ease of use, not the benchmark chart.
Who This Is For
Read if. Skip if.
READ IF YOU ARE…
You track AI model releases weekly and want the actual differences between GPT-6 Astra, Gemini 3.8 Flash, and Fable 5.1 without reading three separate blog posts.
You use Gemini Notebook (formerly NotebookLM) regularly and need to understand the new compute-based usage limits before you hit one mid-project.
You're deciding which model to make your daily driver and care about real-world cost and friction, not just benchmark scores.
SKIP IF…
You already read the GPT-6 Astra, Gemini 3.8, and Fable 5.1 launch posts directly and don't need a recap.
You don't use Grok Bot or Gemini Notebook, since two of the segments are product-specific updates for those tools.
TL;DR
The full version, fast.
OpenAI, Google, and Anthropic all shipped flagship models in the same week. GPT-6 Astra claims the best computer-use and knowledge-work performance ever, but it's rolling out to businesses first and won't be cheap. Gemini 3.8 Flash tops finance and legal benchmarks and is already the model quietly running inside Search, Gmail, and Docs, while a true Gemini Pro refresh remains missing since February. Anthropic's Fable 5.1 promises fewer false-positive safety refusals, more natural writing, and lower cost than Fable 5, though real usage shows no real price drop. Meanwhile Gemini Notebook replaced per-feature caps with a single compute-based budget (5-hour window plus a weekly ceiling), Grok Bot added a plugin that lets bots complete real purchases with approval, and three hikers got stranded after trusting bad AI-generated trail directions. The throughline: capability wins benchmarks, but ease of use and price decide what people actually keep using.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Cold open framing this as one of the biggest AI weeks in a while: new models from Claude, Gemini, and ChatGPT plus disappointing Gemini Notebook news.
00:39 – 01:03
02 · Three big new models
Names the week's three flagship launches: Fable 5.1 (Anthropic), Gemini 3.8 Flash (Google), and GPT-6 Astra (OpenAI), framed as each being best at something rather than best overall.
01:03 – 05:21
03 · GPT-6 Astra
OpenAI's GPT-6 Astra claims the best computer-use performance ever, rolling out to businesses first, with a promo video showing it operating a live desktop. Benchmarks reportedly crush every prior model, but it isn't cheap and OpenAI's president calls it a possible generational leap toward AGI.
05:21 – 06:22
04 · Sponsor: Scrimba Explain
Sponsored segment demoing Scrimba Explain, a tool that turns a question, document, webpage, or codebase into a narrated AI explainer video, including a ChatGPT plugin.
06:22 – 09:38
05 · Gemini 3.8 Flash
Google's third Flash update in about six weeks. Weak on general knowledge-work benchmarks but tops finance and legal benchmarks, is the cheapest model available, and quietly powers Search AI Mode, Gmail, and Docs. Gemini Pro still hasn't been refreshed since February.
09:38 – 14:52
06 · Fable 5.1
Anthropic's updated flagship, pitched as fewer false-positive safety refusals, more natural/less verbose writing, and lower cost than Fable 5 — though the creator notices no real price drop in practice. Covers why Fable 5's hype collapsed after a mid-cycle access pullback let GPT-5.6 win adoption.
14:52 – 20:18
07 · Gemini Notebook usage limits
Gemini Notebook (formerly NotebookLM) moves from per-feature caps to a single compute-based budget: a 5-hour rolling window plus a weekly limit, a new 'Generate Later' off-hours option, 4x/5x paid-tier allotments, watermark removal for paid users, and a possible unannounced Chrome extension spotted in the launch video.
20:18 – 22:12
08 · Grok Bot updates
Grok Bot adds the Link plugin for AI-agent purchases (human approval required per transaction), a share-as-template feature for bots, and a per-bot voice setting hinting at upcoming voice calling.
22:12 – 24:02
09 · Rapid updates: Google Pics and an AI reliability PSA
Quick mention of Google Pics (AI image generation and editing in Workspace) followed by a news story about hikers stranded after Gemini gave them false trail information, closing with a subscribe ask.
Atomic Insights
Lines worth screenshotting.
GPT-6 Astra, Gemini 3.8 Flash, and Fable 5.1 all launched in the same week, each claiming to be the best model at something specific rather than the best at everything.
GPT-6 Astra is rolling out to businesses first, with personal ChatGPT accounts getting access only in the following days.
OpenAI's president called GPT-6 Astra's capability leap something that 'could be defined as AGI' while the model was still restricted to a small group of businesses for cybersecurity review.
Gemini 3.8 Flash scores poorly on general knowledge-work benchmarks but ranks first specifically on finance and legal-workflow benchmarks among publicly available models.
Gemini's Flash tier, not Pro, is the model actually powering Google Search's AI Mode, Gmail's Help Me Write, and Gemini inside Google Docs.
Google has shipped three Flash model updates (3.6, 3.7, 3.8) in roughly six weeks while a true Gemini Pro refresh has been missing since a 3.1 Pro release in February.
Anthropic claims Fable 5.1 cut cybersecurity false positives by about 60% compared to Fable 5, reducing cases where the model mistook normal coding for security work.
Despite Anthropic's claim that Fable 5.1 is cheaper than Fable 5, real-world usage showed no noticeable drop in cost.
Fable 5's hype cycle collapsed within three months of launch after a mid-cycle access pullback and a price hike, and competing model GPT-5.6 filled the adoption gap in the meantime.
Gemini Notebook (formerly NotebookLM) replaced per-feature generation caps with a single compute-based usage budget that resets every 5 hours, plus a separate weekly ceiling.
A single explainer-video generation inside Gemini Notebook can consume roughly 50% of a free account's 5-hour usage window.
Choosing 'Generate Later' inside Gemini Notebook runs the generation off-hours and does not count against your active usage window at all.
Gemini Notebook's paid tiers get 4x (AI Plus) or 5x (Ultra) the free-tier usage allotment, despite Ultra costing roughly 5x more than Plus.
Grok Bot's new Link plugin lets a bot complete real purchases with one-time-use virtual cards, but every purchase still requires explicit human approval before it completes.
Sharing a Grok Bot as a template copies its configuration into another person's account rather than granting access to the same shared bot instance.
Three hikers who planned an entire mountain trip using Gemini received false information, got stranded off-trail overnight, and had to be rescued.
Takeaway
Capability Wins Benchmarks, Ease of Use Wins Adoption
MODEL LANDSCAPE
OpenAI, Anthropic, and Google all shipped flagship models in the same week, but which one becomes someone's daily driver comes down to cost, friction, and task fit, not the top benchmark score.
03GPT-6 Astra
OpenAI's GPT-6 Astra ships to businesses first, targeting computer-use tasks like filling forms, running QA checks, and multi-agent runs that go for hours instead of single prompts.
Read benchmark claims from AI labs skeptically: a president calling a launch a 'generational leap that could be defined as AGI' is marketing language, not an independent evaluation.
All three flagship launches this week claimed to be best at something specific rather than best at everything, so picking a model now means matching the task, not chasing one leaderboard.
05Gemini 3.8 Flash
Gemini 3.8 Flash underperforms on general knowledge-work benchmarks but tops the field specifically for finance and legal tasks, so a 'worse' general model can still be the right pick for a specialized workflow.
Flash, not Pro, is the model actually running inside Google Search's AI Mode, Gmail's Help Me Write, and Google Docs, meaning most people are already using Gemini without knowing which tier.
Google has shipped three Flash updates (3.6, 3.7, 3.8) in about six weeks while its Pro tier has gone unrefreshed since February, revealing where the company is actually investing.
06Fable 5.1
Anthropic's pitch for Fable 5.1 is fewer false-positive safety refusals, more natural writing, and lower cost versus Fable 5, but real-world usage showed no noticeable drop in cost.
A flagship model's hype cycle can collapse even while it's objectively capable: Fable 5 lost mindshare after a mid-cycle access pullback and price hike, and a competitor filled the gap before the fix shipped.
The model people actually adopt daily is usually the one with the fewest workflow friction points, not the one that wins the benchmark chart.
07Gemini Notebook usage limits
Gemini Notebook switched from per-feature generation caps to a single compute-based budget across everything you create, split into a 5-hour rolling window and a weekly ceiling.
'Generate Later' queues a generation to run during off-hours without touching your active usage window, worth using deliberately if you're close to a limit.
Paid tiers get 4x (Plus) or 5x (Ultra) the free usage pool, so weigh your actual generation volume before upgrading on limits alone.
08Grok Bot updates
Grok's Link plugin lets a bot complete real purchases with one-time-use virtual cards, but every transaction still requires explicit human approval before it completes.
Sharing a bot as a template copies its configuration into another person's account rather than granting shared access to the same bot instance.
A per-bot voice setting appeared before the voice-calling feature itself, a preview signal worth watching for tools that quietly ship UI ahead of the feature it configures.
09Rapid updates: Google Pics and an AI reliability PSA
Google Pics is a new AI image generation and precision-editing tool inside Google Workspace, still in limited rollout.
Three hikers who planned a mountain trip using Gemini and got false trail information ended up stranded overnight, a reminder that AI guidance for physically risky decisions still needs a human, expert gut-check.
Glossary
Terms worth knowing.
Computer use
An AI capability where the model directly operates a computer's interface (clicking, typing, navigating apps) to complete tasks, rather than just returning text.
Compute-based usage limit
A usage cap measured by how much processing power a generation actually consumes, rather than a fixed number of uses per feature per day.
AGI (Artificial General Intelligence)
A hypothetical AI system that matches or exceeds human ability across essentially all cognitive tasks, rather than excelling at narrow, specific ones.
Link
A payment tool built for AI agents that lets a bot complete a purchase using a one-time-use virtual card, with the human still approving each transaction.
Diarization
Automatically labeling which speaker said which part of a transcript when multiple people are talking.
“The model that you end up using isn't always the most intelligent model. It's often the model that's just the easiest to use.”
Single-sentence thesis of the whole video, works standalone with zero context.→ TikTok hook↗ Tweet quote
12:02
“It's just as expensive. I don't see really any reduction in the amount of usage it's costing me to use 5.1 versus [Fable].”
A direct, credible myth-check against a company's own pricing claim.→ IG reel cold open↗ Tweet quote
23:04
“So here's a little PSA: use your brain. I wouldn't just rely on AI for something as potentially dangerous as hiking up a mountain.”
Punchy closing warning tied to a concrete real-world consequence.→ newsletter pull-quote↗ Tweet quote
04:51
“The president of OpenAI said that this new model represents a generational leap in capability that could be defined as AGI. That remains to be seen.”
Bold company claim immediately undercut by the creator's own skepticism.→ TikTok hook↗ Tweet quote
The Script
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
17px
This was one of the biggest AI weeks we've seen in a long time. I'm almost tempted to say that too much happened this week. We got new model announcements for Claude, Gemini, and ChatGPT, and these were big models.
We also got some disappointing news about Notebook LM, or Gemini Notebook as it's now called. There's a very interesting new Google AI app. and a lot more.
But don't worry, I'm here to cut through the noise and to talk about what actually matters in the world of AI, just like I do every week inside this weekly AI News Recap. So make sure to subscribe to stay on top of the latest, and now let's get into it. We first have to talk about all the new models that came out this week.
So first, Anthropic announced Fable 5 .1, then Google released Gemini 3 .8 Flash, and most recently, OpenAI announced GPT -6 Astra. And we're going to talk about all these models because quite frankly, I think all of them are the best at something, even if one is not the best at everything. But I want to start with Astra because this could potentially be one of the biggest model releases of not only the year, but of all time.
So here's what you need to know about OpenAI's newest model called Astra. They have said that this is the most intelligent and aligned model in the world and sets a new standard for computer use. browsing, software engineering, cybersecurity, science, and professional work.
In sum, what they are saying is that for computer use and knowledge work, this is hands down the best model that has ever been released. Let's look at the official blog post. And first, check out the website for this blog post.
It is absolutely... I love what they've done with this and the way that it looks. Let's come down though.
They have a video about it. So I want to play a part of it because it really demonstrates some of the impressive things that Astra can do. So what they did in this video was they had a bunch of people talking to ChatGPT and what ChatGPT is doing on the computer is being projected onto this screen here.
And you can see in real time. as ChatGPT is using the computer and how good it is at using the computer in real time. It is incredibly impressive because normally computer use, where an AI takes over a computer for you, can be very slow.
This looks not only faster, but incredibly accurate. The fact that all this is just being done by talking to ChatGPT Live is incredibly impressive. OpenAI also released a bunch of benchmarks for GPT -6, and it seems to crush every single one of the benchmarks.
If you don't care about benchmarks, that's okay, or understand them, that's okay. This just means that on paper... This is the most powerful model that has ever been released.
And in this blog post, they spend a lot of time talking about how good it is at computer use and how good it is at knowledge work, which for me is wonderful to see. I'm not a coder or developer myself. I use AI tools for knowledge work.
I want to be able to create presentations, documents, do research, things like that. And the models are already incredibly capable for that. But the more we can see companies focusing on that and prioritizing that, I think it's better.
I think it's going to lead to much wider adoption of AI tools. And here it seems like OpenAI is leaning into that quite a bit. And so combined that with computer use, it looks like this may be a model that will be able to agentically handle a lot more work for us than any model could in the past.
Now, a few other things to be aware of. First, It is not available for personal accounts as of right now when I'm recording this video.
It's only available for certain businesses and it will roll out for personal accounts over the next few days. So maybe by the time you're watching this, you will have access to Astra. And certainly once I get access to it, I'll be making videos exploring what you can do with it.
So make sure to subscribe if you wanna see my take when I actually get to try out Astra. But the other thing you want to be aware of is that it's not going to be cheap. It is going to be as expensive as Feeble 5 .1, so it may not be the best everyday model.
For that, you may want to use one of OpenAI's other models. So how we end up balancing that is something else I would really like to explore. I'm trying to reserve judgment on this until I actually get to test it myself, but I've spoken to a lot of friends who have had early access to it, and they have said this is the most capable, intelligent, and quite honestly, just mind -blowing model they have ever used.
So I can't help but feel really excited by this. This feels like it may be another big moment for AI that changes everything. In fact, the president of OpenAI, and this is the president of OpenAI saying this, so take it with a grain of salt, but he said that this new model represents a generational leap in capability that could be defined as AGI.
That remains to be seen, but let's see how it is in a few days when we actually get it. Every once in a while, I try out a new AI tool that makes me go, whoa, that's cool. And that's what I thought when I first tried out the sponsor of today's video, Explain by Scrimba.
So check out how this works. Let's say there's a topic that you're curious about. We can ask a question about it right here in the prompt box.
But unlike other AI tools that would just give you a text response or maybe a diagram, this instantly generates for you an AI explainer video. Bluetooth is the invisible thread that connects your phone to your headphones. or your car you see how fast that was it's wild and it doesn't just work with questions you can also have it turn documents web pages or even a code base into a narrated explainer video and the best part is since it has a plugin with chat gpt you can generate these inside your chats without switching tools explain and chat gpt will create visual walkthroughs with narration and diagrams in a way that's easy for you to understand explain is free while it's in open beta so you might as well try it out make sure to use the link below to sign up and now let's get back to the news all right let's talk about Google's newest model Gemini 3 .8 flash and what's interesting about this is that 3 .7 flash just came out three weeks ago and three weeks before that 3 .6 flash came out so this is a pretty fast turnaround and they're updating their flash model
very frequently now so what's new with 3 .8 flash well they released some benchmarks and on this channel i don't like to spend a lot of time talking about benchmarks i want to see how the models actually perform in the tasks that i give them but it is a good indication of what's being promised by these models or for these models so the first benchmark is one that measures how well the models can handle knowledge work And unfortunately, 3 .8 Flash has scored pretty poorly on this.
There are plenty of other publicly available models that perform much better than 3 .8 Flash. However, if we look specifically at finance and how well these models handle financial analyst tasks, you'll see that Gemini 3 .8 Flash is the top performing model. And if you look at legal work, you'll see that 3 .8 Flash is the top performing model.
And what's perhaps the most interesting is that 3 .8 Flash is the cheapest model by far, which means that you can get a lot more done with 3 .8 Flash than you can with a lot of other models for the same price. Now, as for availability, it is currently available already inside the Gemini app. You can see it right here, 3 .8 Flash.
And this is the model that you should be using because it is the best model available. from Gemini, way more powerful than their Pro model, which is still on 3 .1 Pro. Speaking of which, there's still no sign of their new Gemini Pro model.
3 .1 Pro came out in February. We had an announcement at Google I .O. in May that 3 .5 Pro was supposed to come out, and it never came out.
We had several Flash models come out in the meantime, and still no Pro model. so that upsets a lot of people but here's what most people get wrong the average consumer has no idea what gemini model we're on they have no idea what the benchmarks are and they probably don't care about the benchmarks a lot of times people don't even know they are using gemini or ai at all instead they interact with ai inside of tools like google search when they use ai mode or inside of Google Docs when they use Gemini down here, or inside of Gmail with the Help Me Write feature.
And guess what? All of those are powered by Flash. They're not gonna use Pro for that because it takes too long.
So Google needs a really powerful model that's really fast and really cheap so that it can be used inside of all of their products. So it's not super surprising that they're focusing so much on the Flash model. Now, That doesn't excuse them entirely.
They definitely need to catch up and finally release a pro model, but I don't think it matters as much as a lot of critics say. And finally, Anthropic released Fable 5 .1. And I think it's kind of ironic that I'm talking about this model last, because in the past, this would have been the most exciting release.
We were really excited when the first Fable 5 came out. We'll talk more about sort of that. aspect of it in a second.
But first, let's actually talk about the model itself and what was released. So these are the benchmarks for Fable 5 .1. And we'll just talk about the ones that actually matter.
So first, knowledge work. It is really good at knowledge work. Now, I don't have anything comparing Fable 5 .1 against Astra because both these models are still too new.
But compared to Fable 5, Cloud Opus 5, and GPT 5 .6, It is the best model for knowledge work. Anthropic is also making a lot of other claims about the new 5 .1.
So one of the reasons that a lot of people stopped using Fable, even though it was such... a powerful and intelligent model was because it had a lot of shortcomings, I would call them, that made it very difficult to work with. So first, the original Fable 5 had a lot of guardrails to make sure that we weren't using it improperly.
And that resulted in a lot of false positives. Anthropica said that new 5 .1 will have fewer false positives. And in the limited time that I've been able to use it, it does seem like that is the case.
Another problem with Fable 5 was that it tended to be very verbose. And at times I would be speaking to it or speaking to it, and I would feel like dumb. Like I wouldn't understand what it was trying to tell me.
And I would always keep telling it, can you dumb that down for me? And now this version 5 .1 should have more natural language and be able to communicate in a way that makes a lot more sense for a human. And again, this is something I've already noticed about the new 5 .1.
And the final important thing is that Anthropic has said that 5 .1 will be cheaper. This was a big complaint about Fable 5, was that it was just too expensive to use as an everyday model. So you'd only use it for your most important tasks or to plan out your tasks, and then you would give the tasks to Opus or maybe even ChatGBT to finish it up.
Now, that's what Anthropic has been claiming. However, from what I have seen, it's just as expensive i don't see really any reduction in the amount of usage it's costing me to use 5 .1 versus vibe maybe there's a difference but it's so little it's really hard to notice But overall, it is definitely a better model, a lot easier to use, and a lot nicer to use than Fable 5.
Again, we went over the cost. It still feels just as expensive. Now, the next thing I want to talk about is sort of the hype cycle around Fable 5 and why I don't think more people are talking about the new 5 .1.
And it really comes down to the way... all this was rolled out so when fable 5 first came out the hype was so high it was one of the most exciting model releases of the year and then it just went through this entire roller coaster because shortly after that the government removed access to fable 5 so we couldn't use it and then it came back but anthropic said you will get access to fable 5 but It will only be part of your subscription for a few days.
And then after that, you have to pay more to access it. And they kept extending that deadline. And during that time period, GPT 5 .6 came out.
And we all tried GPT 5 .6 and we're like, this is a fantastic model. It's really easy to use. It doesn't have maybe the intelligence of Fable in all aspects, but it is so much easier to use.
And it doesn't have a lot of the problems that we were facing with Fable. talks a lot more naturally it lets us use it it doesn't throw up all these guard rails and it's not super expensive like fable so a lot of people just said i'm not going to use fable it's just not worth the hassle when gpt 5 .6 works so well so even when fable 5 went onto our subscriptions and we didn't have to pay extra for it a lot of people just didn't switch over to using it because 5 .6 made a lot more sense.
And I think that entire sort of timeline and experience really killed a lot of the hype for Fable in general. So this new Fable 5 .1, not as many people are excited by it as they were excited about Fable 5. I tell you all this to make a point, which is that the model that you end up using isn't always the most intelligent model.
it's often the model that's just the easiest to use so for myself over the next few days i'm certainly going to be testing the intelligence of astra and fable 5 .1 but i also want to see how well and how easily we can use those models compared to something like gpt 5 .6 which is honestly one of my favorite models to work with so that's what i'm going to be looking for over the next few days as i'm trying out all these different models Now on to some rather disappointing news.
If you're a fan of Notebook LM, or as it's now called Gemini Notebook, you may have noticed that notebooks now have usage limits. That's right, and this was a little bit unexpected. I mean, maybe it shouldn't have been.
They can't just give out compute for free forever, but it did just kind of seem like one of those things that we... were taken for granted and never really saw coming. So here's the deal.
I got this email last week kind of out of the blue, and it said that Notebook LM is moving towards a compute -based usage limit. So in the past, there were limits on using Notebook LM, but they were for features. So opening up on these notebooks, for instance, we have video overviews, for instance.
You could generate video overviews in the past, of course. but you could only generate a certain number per day and now they're moving away from that and moving towards compute based usage limits that apply to everything that you're creating inside of notebook lm and it's not like a set number of generations you can do for each feature like we had in the past so what does this look like well inside of gemini notebook now if you click on settings on the top right you will now see an option for usage.
And if you select that, you'll be able to see your usage limits. There are two limits. The first one is the current session, and the second one is the weekly limit.
If either of those hit 100%, you're not gonna be able to use Gemini Notebook until it resets. So the Kind of good news here is that in the past, the limits that were on your account, like the number of videos you could generate or the number of images you could generate, would only reset once per day.
But with the new usage limits, those reset every five hours. But we do now also have a weekly limit as well. So realistically, what does that mean well i can only show you this inside my free google ai account because i don't yet have usage limits inside my paid ones i know it is coming but i don't have it yet so inside this free account though what i can show you is if you open up any of the asset generators like let's say you want to generate a video you'll notice down here there's a little bar and this will show you how much you've already used in your five hour window which is zero percent right now and a prediction for how much this generation is going to cost you.
And so for this one video, it looks like it costs more than 30 % of that five hour usage window. And that's for a short video. If I choose an explainer video, it looks like it jumps up closer to 50%.
What I thought was pretty interesting was that if I select slide decks, you'll notice that uses It looks like over 50 % and it doesn't make a difference whether I choose the default length or a short length. But what they've also added, which is really nice, is if you are getting towards the end of your usage limit for your five hour window, you can always click on generate later instead of clicking on generate now.
That will not count against your current limit. It will instead start the generation later when you're not using Gemini Notebook. So it won't count against a five -hour window where you're actually working.
So for instance, you could set this right before you go to sleep. And then when you wake up in the morning, your video or your slide deck will be complete. And it didn't count against a usage window when you were actually on your computer and working.
So that's kind of nice. They've also said that for a pro subscription, you will get 4X the usage of a free account. And for an Ultra subscription, you will get 5x the usage, which doesn't really make a lot of sense to me because Pro costs $20 a month and Ultra costs at least $100 a month.
So I don't quite understand that math. But the point is, with a paid subscription, you will get more usage than you get on a free subscription. Oh, and two more updates about Gemini Notebook.
So first, if you come over to Settings now, you will see that there is an option to remove watermarks. Let me actually go over to my paid account. Here we go.
So this is my paid account. And for paid users, you can remove watermarks now. This was something that was rolled out for Gemini and now it's available for Gemini Notebook.
So if you generate images and videos, you won't see the watermarks on them as long as you have this checked off where it says remove watermarks. The other thing is a bit of a rumor. So this right here is the video that the Gemini Notebook team released when they announced all these new usage limits.
And let me know if you see something interesting in here, particularly right here. You notice that inside the settings when they're highlighting how you can select your usage, notice what's above that Chrome extension. that's something we've never seen before so it seems like at least internally the gemini notebook team is testing some kind of chrome extension for gemini notebook so we'll definitely keep our eyes on that i don't think this was an intentional leak but there it is it seems like that we may get a native chrome extension for gemini notebook Next, some quick announcements about Grokbot.
So first, they added a new plugin called Link. This allows Grokbot to actually complete purchases for you using one -time use credit cards. So if you want your bot to actually go out and make purchases for you, it can now do that with Link.
If you're not familiar with Link, it is a tool that is specifically designed for AI agents to be able to make purchases for you. And what's nice about it is that you still have to approve every purchase. It can't just go out and make purchases on its own.
You will get a notification asking you if you want to allow the purchase, and then you can verify it. But the rest of the process, like the bot, actually searching for a product and going through different comparisons and then choosing one, adding it to the cart and checking out, all of that can be done by the bot and you just approve the final transaction.
The other thing you'll now notice in GrokBot, after you open up any of your messages, threads that you have with your bots, click on their name on the top. And then on the right here at the bottom, you can now share your bot as a template.
So if you have a bot that you really like and you want someone else to be able to use it, you can just share it. It doesn't mean you are sharing in the sense that you're both using the same bot. It just makes a copy of it inside their account.
Another thing I just noticed here is that I now have the option to choose the voice for each of my bots. And this, was a rumor I had heard about a few days ago where we will soon get the ability to actually call our bots, which is basically like voice live, but that is not available yet.
There's dictation, but there's no voice calling, but it seems like it definitely is coming because we have the option to choose the voice for our bots now. Google is also rolling out access to a new tool called Google Pics. I don't have access to it yet, but I'm mentioning it because I got a preview of this at Google IO in May, and my mind was blown by what this can do.
If you do any type of image generation, you will definitely want to see what this tool can do. So as soon as I get access to it, I'm going to make a dedicated video about it. If you don't want to miss out on that, make sure to subscribe to the channel and turn on notifications so you get a notification as soon as this video drops, hopefully in a couple of days.
And the final story is about three hikers in California who tried to hike a mountain. They started later than they should have. They didn't bring enough food or water.
At a certain point, they realized they were in trouble because it was getting dark. They called for help, but then they wandered off trail and then ended up spending the night on the mountain. unplanned well after they got rescued they revealed that they had planned their entire trip using google gemini and it gave them a lot of false information so here's a little psa use your brain i wouldn't just rely on ai for something as potentially dangerous as hiking up a mountain you definitely want to be smart you definitely want to see what experts have to say about it.
I don't think there's anything wrong with using AI to guide you in the right direction. But remember, AI can still make mistakes and does. So quite the crazy week in the world of AI.
It will be interesting to see over the next couple of weeks. how these new models end up working and which one we end up actually using day to day. But that's all I got for you for now.
Remember, if you enjoyed this, consider subscribing. I'm closing in on half a million subscribers. So if you want to help me out there, subscribe and then I'll see you in the next one.
Bye for now.
The Hook
The bait, then the rug-pull.
Three labs, one week: OpenAI, Google, and Anthropic all shipped flagship models within days of each other, and the creator cuts through the announcement-blog hype to find out which of them actually matters once you have to use it every day.
Frameworks
Named ideas worth stealing.
03:40list
GPT-6 Astra's headline claims
Businesses first, personal accounts later
Destroys almost every benchmark
The most intelligent model in the world
Insane computer use
Built for agentic, multi-hour work
Expensive, in Fable 5.1's price range
Six claims OpenAI made about GPT-6 Astra at launch, as pulled from the announcement blog and benchmark posts.
Steal forA template for structuring a skeptical model-launch breakdown into claim-by-claim cards.
11:05list
What Anthropic says makes Fable 5.1 better
~60% fewer cybersecurity false positives
Finds bugs without exploiting them
More natural, less verbose writing
Cheaper at low and medium effort
Fewer fallbacks on basic biology safeguards
Enterprise frontier safeguards
The six day-to-day improvements Anthropic claims for Fable 5.1 over Fable 5, beyond the benchmark charts.
Steal forA checklist for evaluating whether a model update actually fixes daily friction, not just benchmark scores.
CTA Breakdown
How they asked for the click.
VERBAL ASK
05:21product
“check out the sponsor of today's video, Explain by Scrimba... make sure to use the link below to sign up”
Live in-app demo of the sponsor's product (typing a real question, playing a generated explainer video) rather than a read-only ad slate, then a direct link-in-description ask.
Add Modern Creator as a preferred source and Google shows you more of our breakdowns in Search, Top Stories, and AI Overviews. It only changes what you see, and you can undo it in your Google settings anytime.
Add to Preferred SourcesOpens your Google source preferences with us pre-loaded. Tick the box and you're done.
A YouTuber walks through the eleven-bot staff living inside Grok Bot, each with a name, a job, and a cloud computer that keeps working after he closes his laptop.
A YouTube tutorial on wiring ChatGPT and Claude into a $79 desktop editor called Borumi so they cut, zoom, and animate your talking-head videos for you.
A tour of ChatGPT's built-in site builder — from a class homework page to a private morning-briefing dashboard — built from nothing but plain-English prompts.
A launch-day test of Anthropic's new default Opus model — benchmarks, a landing page build, a script draft, and a motion graphic, all run head-to-head against Fable 5.
00:21
01:03
01:21
01:43
02:15
02:46
03:09
03:27
03:40
03:55
04:21
04:39
05:04
05:21
05:56
06:09
06:45
07:03
07:21
07:55
08:13
08:33
08:55
09:09
09:18
10:03
10:21
10:39
10:57
11:15
11:33
11:51
12:10
12:28
12:46
13:22
13:40
13:58
14:28
14:52
15:19
15:31
15:47
16:00
16:22
16:40
16:58
17:16
17:34
17:52
18:10
18:28
18:46
19:01
19:33
19:58
20:17
20:38
20:52
21:13
21:28
21:46
22:12
22:22
22:58
23:16
23:34
13:02
15:58
13:35
12:18
14:08
07:18