Claude Fable 5.5 is Coming + More (AI News)

0:00 Another week in more AI updates and a few of these releases might change how you use AI every single day. And we've also [clears throat] got some big news on Claude Fable 5.5, which looks like it

0:11 could be coming as soon as next week. So, I'll tell you everything we know so far. Google just put out a new Frontier model inside of Gemini. And Gemini can now go and do tasks for you even when

0:21 you're not at your computer. Your Grockbot can now have its very own email address. And there's a new AI voice model that sounds so real. I'm genuinely blown away by it. And Meta just open

0:31 sourced one of its best AI models, and they actually teamed up with Apple, so you could run it right on your own desk. So, there are five big releases this week that I want to cover. And I'm going

0:40 to break down what changed with each one and test a few of them out. And honestly, AI is moving so fast right now that staying on top of all of this could be someone's entire job. That's exactly

0:50 why I put out these episodes every single Friday so you could get everything that actually matters to you in one quick video without spending the whole week digging through everything

0:58 yourself. So without further ado, let's dive right in. Google just released its brand new Frontier model called Gemini 4 Argon. It also added a new way to have Gemini go and do tasks for you. Now

1:09 Google first announced it at the end of September, but only a small group of security partners could use it. And now it's open to anyone paying for Google AI Pro or Ultra, both inside of the Gemini

1:20 app and inside of AI mode in Google search. So the question is, what is actually different about it? Google says Argon answers more directly with less hedging. So basically, you're going to

1:30 get a few of those answers where it says it depends five different times before it actually tells you anything. Google also says it's a lot better at coding and keeps you updated on its progress

1:40 during longer tasks. And in Google's own posts, you can see some pretty examples right here, like one where it learned how to throw a cornhole bag just by watching a video and then a humanoid

1:50 robot made seven shots in a row. But I'm honestly more excited about the other part of this release, which is a new feature called tasks. And this is a feature that if you are a pre-existing

2:00 Gemini user, you are going to love. So, right inside of the main chat window, we now have a switch that flips Gemini from just chatting to actually doing tasks for you. And Google says Gemini can now

2:12 work for you even when you're not at your computer across your Google apps and over 40 others like Instacart and Open Table. Google's own examples are things like ordering your groceries or

2:21 booking a table at a restaurant. And if that sounds familiar, it's because this is basically Google's version of Claude Co-work where Claude goes and does tasks for you across your different apps and

2:31 keeps working even when your computer's closed. So, this isn't anything that's revolutionary or groundbreaking, but we can now access this directly inside of the Gemini ecosystem. And I think this

2:41 is probably Google's foot in the door to building a full personal AI agent like Grockbot or Metamuse, Chachabit Dots, and all these personal agents all these companies are building. It honestly

2:51 feels like Google's been on the sidelines for a while, but maybe this is a start of them getting back into the AI race. Now, there is one catch with this release, and that is who actually gets

3:00 access to tasks. Google's announcement says it's for pro and ultra, but on Google's own plans page, it only lists it under ultra in select countries. Personally, I'm going to stick with

3:10 Grockbot, but I genuinely think that Google has a huge advantage here because so many people already live inside of all the Google application. Next up, Grockbot just gave your bots their very

3:19 own email address. And this is a major upgrade that genuinely changes how we're going to use our AI agents. This is a big step toward truly autonomous AI agents. And I'm going to explain exactly

3:30 why. Your Grockbot no longer has to work out of your own inbox so it could sign up for services, contact businesses, or book time with someone all on its own. The simple way to think about this is

3:40 it's like hiring an assistant and giving them their own work email instead of handing them the password to yours. For example, I could give my brand partnership bot its own email address

3:49 and hand that out to brands and it reads those emails and works on them for me without ever touching my personal inbox. I've already set this up inside of one of my own Grock bots. Now, one thing I

3:59 do want to point out is that it can't just go and send emails on its own if you don't want it to. XAI says any email it sends from that address awaits for your approval on a draft card. So,

4:09 you're going to see the recipients and the message right here, and then you choose to send it or discard it. Next up, we have some news about Fable 5.5 potentially releasing next week. Now,

4:20 when Fable 5.1 dropped back in September, it was already an incredibly powerful model. But since then, we've seen the entire Claude 5.5 family come out. And honestly, these models are

4:31 incredibly good. Claude Opus 5.5 is by far the best model I've personally ever used. And that's exactly why I'm so excited about this because it set the bar very high. And since Fable is the

4:42 most intelligent model in the Clawude family, odds are Fable 5.5 is going to be incredibly powerful and capable. It showed up on October 13th in Microsoft's list of models on Azure, which is one of

4:55 the places businesses use Claude, and it was marked as generally available. And Fable is now the only one of the 5.5 models that isn't out yet since Haiku 5.5 just came out on October 7th. So, it

5:07 really does feel like this is the next model to drop. Now, to be clear, Enthropic hasn't announced anything yet. So, there's no official release date, no price, or any benchmarks. But when a

5:16 model starts showing up in a place like this, it usually means a release isn't that far away. And honestly, I'm very excited about this one because I think Claude has made a huge comeback since

5:26 the 5.5 models came out. And I'm actually using Opus 5.5 to edit my entire YouTube videos for me from beginning to end, including all the graphics and animations that you're

5:36 seeing in this video throughout. So, if Fable 5.5 is a step up from that, I genuinely can't wait to see what it could do. And one thing that's really impressed me about Opus 5.5 is how

5:46 efficient it is with usage. I'm on the $200 a month plan and I'm using it constantly for really complex work like this. And I basically never even use half of my weekly limit. Now, just to

5:57 give you an idea on the cost difference right now, Opus 5.5 costs $4 per million input tokens and $20 per million output tokens, while Fable 5.1 costs $10 going in and $50 coming out. So, Fable is 2

6:11 and a half times more expensive. So, the big thing I'm curious about is whether the jump in capability with Fable 5.5 is going to be worth the extra cost, especially since Fable models use up

6:20 your limits a lot faster, and Opus 5.5 can already do pretty much everything I need it to. Now, as soon as this comes out, I'm going to be making a video breaking it down. So, make sure that

6:30 you're subscribed so you don't miss that. Next up, Fish Audio just released a new AI voice model called Drama 3. And honestly, it takes AI voice to a completely new level. And I'm genuinely

6:40 blown away at how good it sounds. If you've never heard of Fish Audio before, they make AI voices. So, you could basically type in text and it reads it out loud in a voice that sounds like a

6:50 real person. I actually use Fish Audio myself for some of my video voiceovers. Now, Fish Audio says this is their most emotional voice model yet with far more feeling than their last one. And the way

7:01 it works is that you just tell it how you want each line to feel in your own words, the same way you direct a real actor. And here's the part that is a really big deal. Drama 3 is open weight.

7:12 That basically means that Fish Audio released the actual model for free. So, anyone with a powerful enough computer can download it and run it themselves instead of being able to use it only

7:23 through Fish Audio's website. With an open model, developers can build their own voice apps on top of it. You're not stuck paying per use, and your audio never has to leave your own computer.

7:33 Now, as far as who gets it, it's live right now inside of Fish Audio's text to speech, their iPhone app, and their API. Now, I want to actually show you how this works with my own voice. And you

7:42 can see that you actually have three options right here. Instant voice clone is where you upload or record a short clip and it clones it right away. Voice design is where you just describe a

7:52 voice in plain text and it makes one up for you. And professional voice cloning is a verified studio quality version that takes an hour or two worth of audio. For this, I'm just going to go

8:02 with instant voice clone because it only takes about a minute. From here, you could either upload a file or record yourself right on this page. And it takes pretty much any type of audio or

8:12 video file. So, what I'm going to do is just upload about a 30 second clip of my own voice. And this is what the reference audio sounds like. Let's say I have a meeting with someone on my team

8:23 or I'm even just doing a voice brain dump of everything I need to get done over the next week. And you can see right here that it says 30 seconds is enough for most voices, but giving it

8:32 more audio can always help. Once it analyzes the audio, it actually fills in all of the details for you. So, it wrote a description of my voice, picked a name, and tagged it as a young male and

8:43 good for narration in social media. And you can change any of these if you want. Now, this part is important because you get to choose who can use your voice. Public means anyone on fish audio can

8:53 find it and use it. Unlisted means that it won't show up in search, but you can share it with someone using a link. And private means only you can use it. Personally, I'm just going to set mine

9:03 to private. And Fish Audio actually makes you confirm that you have permission to clone the voice, which makes sense. So, you can't just clone anyone's voice without their consent.

9:11 I'm just going to rename mine to Brock Voice. And just like that, I have my own voice clone. And here's exactly what it sounds like. Our new platform streamlines your entire workflow by

9:22 automatically organizing your meeting notes and action items. Now, in order to actually use it with drama 3, I go over to text to speech, pick my voice on the right side, and then just pick the model

9:33 I want to use. And you could see that drama 3 is right at the top. Then, if I open up the tags, this is where you tell it how to say things. So, up top, you have tags that only change the words you

9:44 highlight, like whisper, emphasis, faster, or slower. And then below you have emotions like angry, sad, or excited, and even sound effects like laughing or clearing your throat. And

9:54 you can write these directions in English or whatever language you use, but the easiest way is just to hit auto tag and it reads your script and adds a directions for you. So right here, it

10:04 added a clear, confident, persuasive tone and put emphasis on the most important words. And here's exactly what it sounds like. Our new platform streamlines your entire workflow by

10:16 automatically organizing your meeting notes and action items. And I could still add my own tags on top of that, like a pause right here. So that way we get full control over exactly how it

10:25 reads your script. And this is where it actually gets really interesting because I'm personally using fish audio to make short form videos for me on autopilot. So basically I have a system that scans

10:34 GitHub and everything that's trending in the AI space every morning and it finds the best ideas for that day. Then Claude writes a script for each one in my style and it sends that script to Fish Audio's

10:45 API which reads it out loud in my clone voice. From there, my editing system builds the whole video for me with all the animations, captions, and the music, and then sends it to me over email every

10:56 single day. All I have to do is reply to that email, approve the ones I like, then it posts them for me automatically. And now that Drama 3 is out, I just switched my entire system over to it.

11:05 So, I'm going to show you how to connect Fish Audio to Claude Code or any of these other AI platforms yourself. I'm inside of Fish Audio. I'm just going to click on where it says developer, and

11:14 this is where you set up the API, which basically just means you pay as you go. post. You could only pay for what you actually generate. Then click on create API key. Give it a name and pick when it

11:24 expires. And I personally just selected 7 days. It only shows you the key once. So you're going to need to copy it and save it somewhere safe and never share this with anybody. Next, I'm going to

11:33 come over to claude code and just say help me implement my fish audio API into ourv file. The env file is basically a private file on your computer where all these different API keys live. And you

11:45 could see right here that Claude Code made a spot for the key and told me to paste it in myself. Then I just need to go back to my voices inside of Fish Audio, click on the three dots next to

11:54 my voice and copy the model ID. I paste that inside of Claude Code so it knows exactly which voice to use. And now that it's all connected, I'm just going to say create a faceless short form video

12:04 with my new voice and recreate it with this specific short form video. And here is the output that I made. And it's pretty crazy. This free GitHub repo lets you reverse engineer and recreate any

12:15 software you want. It's called REA, which is short for reverse engineer anything, and it already has over 130,000 stars on GitHub. If you guys want to try out Fish Audio, there will

12:26 be a link in the description to sign up and test it out yourself. Next up, Meta just announced Muse Wonder, which they are officially calling its new Frontier model, and it's a pretty big deal. That

12:37 basically means that Meta says they're now up there with the best models in the world like Claude and ChateBT and it was built specifically for a personal AI agent kind of like Grockbot or Chachabt

12:48 dots. Now moving on to the part that everybody wants to see and that is how it compares to other top models. So here's the table from Meta and they're comparing both Wonder models against

12:58 Claude Fable 5.1, Claude Opus 5.5 and GPT6 Astra. And GPT6 Astra is the model that chatbt Dots run on and Opus 5.5 is one of the models that Grockbot can now use. So this is pretty much Meta's new

13:13 model against the other models behind the other two AI agents. The first thing that stands out to me is legal work because on this legal agent test, Wonder Full scored 13.1%

13:24 while GPT6 Astra got 8.6% and Opus 5.5 got 4.2. Now those numbers are low across the board because the work has to meet every single requirement to count but wonderf full is still about three

13:37 times higher than Opus here. Now to be fair it doesn't win on everything and Meta shows that right inside of the tables from this release article on the test where the model has to build an

13:46 entire program from scratch. Opus 5.5 got 70.5% and Wonder Full got 59 and both Opus and Fable are still ahead on computer use. Now moving on to two features that Meta is really pushing

14:00 with this model. And the first one is called active memory. So basically Meta says the model carries the important context about you across your different conversations and then it actively pulls

14:11 up the details that matter for whatever you're working on. Now the second big thing here that is different from previous models is called active focus. And this is all about what the model

14:20 does when a task gets really hard. Meta says it keeps working toward your goal by trying a different approach, working through whatever is blocking it, and adapting when new information comes in.

14:31 It also asks you for clarification when it needs it. And Meta says it checks in with you before it takes any important action, which is exactly what you want from a model that is acting on your

14:40 behalf. All right, moving on to price. And there's a catch here that I think most people are going to miss. For developers using the API, Muse Wonder is $4 per million input tokens and $20 per

14:52 million output tokens. That's the exact price as Opus 5.5, and it's a lot cheaper than Fable 5.1, which is $10 and $50. It also has a 1 million token context window, and Meta doesn't charge

15:04 extra for long prompts. Now, here's the catch, and this is for the model called Wonder Full. As of right now, it's not available through the API at all. The only way that you can get access to this

15:14 is through Muse Code, which is Meta's coding tool, and it's only on the $50 a month or the $200 a month plan. All right, so who can actually use this today? As of right now, Muse Wonder is

15:25 live in the MetaMo API and inside of Muse Code and Wonderful is live for Muse Code subscribers on those two higher plans. Now, if you're not technical like me and you just want this inside of the

15:35 Muse app, it's not inside of there yet. Meta says Muse Wonder will soon power experiences inside of Muse and it's going to roll out to paying subscribers first, but they didn't give an exact

15:46 date. So, as of right now, I can't tell you how it compares to Grockbot day-to-day because the only way to really know is to use it inside of Muse. And Meta wasn't done with releases for

15:56 the week because they also just open sourced one of their best AI models. So, you could download it for free and run it right on your own desk. It's called Muse Spark Open. And Meta says it's the

16:06 most capable American openweight model yet. And just like we talked about earlier with Fish Audio, OpenWeight basically means that Meta released the actual model itself. So anyone with a

16:16 powerful enough computer can download it and run it themselves instead of only using it right inside of Meta's own apps. Now, the part that I think is really interesting is that Meta actually

16:26 teamed up with Apple on this, which is kind of surprising because these two companies don't usually work together. And you can see right here, it's running on two of the new Mac Studios connected

16:36 together. But I'm going to be honest with you, you're going to be spending well over $20,000 in order to run it at home, exactly like this setup right here. But genuinely, I think this is a

16:45 look into the future of how we're going to be using AI. Right now, pretty much everything we do with AI runs on some company's servers, so you have to pay every single time that you use it. Along

16:54 with that, they have all of our information, but with a local model, it's running on your actual computer, so you own all this information. So that way your files never leave your computer

17:03 and there's no bill for every message or paying monthly subscriptions. I know $20,000 is a lot of money today, but I think this kind of setup is only going to keep getting cheaper, especially once

17:12 better models are more accessible via open source. Meta already has a smaller model called Muse Glimmer that runs on a computer. I think right now the easiest way for most of us to try Muse Spark

17:22 Open is through a site like Open Router without having to buy any of that hardware. And the reason this is such a big deal is that most of the best open- source models over the last year have

17:32 been coming out of China. For example, Kimmy and Quinn. So, this is basically America's answer to that. Now, to be fair, it doesn't win on everything. In Meta's own comparison, Kimmy K3 still

17:42 beats it on a couple of the coding tests, but personally, I genuinely think this is a really good thing for all of us because having more good open source models is what keeps pushing the price

17:52 of AI down. And the last big thing this week is that people are now using AI agents to rebuild entire apps from scratch, including Photoshop. One developer used Claude Opus 5.5 to

18:03 rebuild seven of Adobe's apps, including Photoshop, Illustrator, and Premiere. And he released all of them free as open- source. Adobe has been building Photoshop for over 30 years, and one

18:15 person just rebuilt an early version of it with AI. To be fair, he did say it's still early and not ready to replace Photoshop for professional work yet. And he built it without any of Adobe's

18:25 actual code. And it's not just Adobe because there is also a free GitHub repo called REA, which is short for reverse engineer Anything. And it plugs into tools like Claude Code and Codeex, and

18:37 it lets your AI open up an app you don't have the code for and figure out exactly how it works. This GitHub repo is now already at over 224,000 stars. Now, the question is, why does this even matter?

18:49 I personally think this is genuinely one of the clearest signs of how fast AI is moving because one person with a few AI agents is now doing what used to take a whole team of experts. And the way to

18:59 think about it is if anyone can now figure out how an app works and rebuild it, the software alone won't protect a company anymore. This is just one of the first tangible examples of how much AI

19:10 is going to change businesses and specifically software businesses in the future. Anyways, that is it for this week's AI news episode. So, make sure you're subscribed so you don't miss next

19:18 week's episode. Thank you for saying the end. I appreciate you and I'll see you in the next video. Cheers.