Getting the transcript
Reading the captions from YouTube. A video nobody has opened here before takes 10 to 30 seconds; this page fills in on its own.
Getting the transcript
Reading the captions from YouTube. A video nobody has opened here before takes 10 to 30 seconds; this page fills in on its own.

TheAIGRID · @TheAiGrid
Words
2,978
Runtime
12:31
Speaking pace
238wpm
Reading time
12min
238 words per minute, above the 201 75th percentile of 349 measured videos. That distribution comes from the 349-video hook study.
Opening (first 30 seconds)
Claude Opus 5 is genuinely very impressive and in today's video we need to talk a lot about what on earth is going on. So I don't want to waste your time too much with this video. I just want to get straight into things because this is a release that I think most people will miss. And the reason I say that is because 90% of the time when there is a model jump increase in capabilities. It can seem like that there is just an incremental jump because you're going from Claude Opus, you know, 5 to 5.5 or 5.4 to 5.5. But I will say that this is not just an incremental jump. This is a
119 words, the words spoken in the first 30 seconds at 238 words per minute.
Free, no signup. See how the first 30 seconds hold attention, with rewrites.
Sentence shape
| Measure | This transcript |
|---|---|
| Sentences | 157 |
| Average words per sentence | 19.0 |
| Longest sentence | 119 words |
| Questions asked | 5 |
What this transcript is
Every word below is the caption track YouTube publishes for this video, pulled from the video itself and reproduced unchanged. It is not Prepublish's writing, not a summary, and not a re-transcription: it is the video's own published captions. English captions, generated automatically by YouTube, in the video’s original language. Source: the video on YouTube. A channel that would rather this page did not exist can ask for its removal through the contact page, and it is removed.
No Script X-ray for this video: YouTube shows a Most replayed graph only once a video has enough views.
Claude Opus 5 is genuinely very impressive and in today's video we need to talk a lot about what on earth is going on. So I don't want to waste your time too much with this video. I just want to get straight into things because this is a release that I think most people will miss. And the reason I say that is because 90% of the time when there is a model jump increase in capabilities. It can seem like that there is just an incremental jump because you're going from Claude Opus, you know, 5 to 5.5 or 5.4 to 5.5.
But I will say that this is not just an incremental jump. This is a huge jump. What we can see here is one of the first times where we have a jump that is confusing. And I say confusing in the sense that it's so good, but it shouldn't be this good. As in, what on earth is going on? How have you progressed this much? We need to understand what on earth is going on. So basically across every benchmark, this is allegedly [music] smarter than GPT6 Astra which very recently was known for doing a lot of crazy stuff such as solving the Navia Stokes problem and other you know Millennium Prize problems as well as making progress on many other you know real research problems.
[music] So the thing here is that how on earth did Claude just go ahead and update Opus 5.5 [music] and then across every benchmark such as agentic coding you know such as knowledge work such as you know multi-disiplinary reasoning um and computer use and visual chart recognition all of these benchmarks how is it not only better than GBC Astra but also Fable 5.1 because I would have presumed that the Fable class level of models are larger you know they're probably like a bigger model and thus have more compute behind it and thus if you expected gains you would expect it to come from maybe Fable 5.5, but this clearly is an Opus level model.
And not only that, okay, not only is it so much more intelligent, another thing that most people are going to miss from this these benchmarks is that the pricing is that it's actually quite cheaper. Now, of course, it's not as cheap as, you know, the GPT series of models or other models. It's still expensive when it comes to Clawude models cuz you know how much they talk. But you can see here that it's actually a bit cheaper.
And of course, it might not seem like much, but over time, this will add up. So you can see that Claude Opus costs $4 per million tokens for input whereas Opus 5 is $5 and the Opus tok and the output tokens are $20 versus $25. So this is something that I think is actually important because the model is smarter and it is cheaper which means that they probably behind the scenes made some kind of breakthrough. Now one thing I will say okay before we get into all of the crazy examples is that there is some quality of life changes.
One of the reasons I and many other people hated using the clawed models is because the way they spoke was just absolutely awful. It was like talking to I'm not even sure how to describe it, but it was just like, "What on earth are you saying in this response? This isn't even English. This is just like a super confusing response." And this was something that would essentially ruin the quality of life in terms of chats.
Now, OpenAI didn't really have that problem, but the fact that OpenAI and other companies are starting to see that this is a big thing shows us that they've actually managed to change this. Look at this feedback. And I think Opus now is in that stage where look, it is a model that is usable again. So, this is something that is very, very impressive. They've listened, they've made it cheaper, faster, better, and they've also made a quality of life change where the responses are a lot more legible and it's very, very easier to understand what's going on.
I think there was an issue where they trained the model and because they trained it in a for aentic work, it just reasoned and those reasoning output tokens just got put into the final response and that's not really what we wanted to see. We want to see a real final response. Um, and some people are basically saying that the previous version of Opus 5 was just meant to be AIAI conversation because they were essentially sub agents for Fable.
Either way, we have a really optimized Opus 5.5 and it does seem like right now this might be the smartest model in the world. Now that there is some caveats because depending on your workflow, OpenAI going to be better maybe in terms of mathematics and of course they did leave out certain benchmark areas where GPT6 Astra is going to be performing better. So as always do test this but when it does come to coding it does seems like this Opus 5.5 model may have the edge.
And while all these new AI tools are becoming so much more powerful the annoying part is that most of them are locked behind completely separate subscriptions. And that is why today's sponsor chatm by Abacus AAI is worth checking out. You see, Chat LLM brings Claude, [music] Chat GBT, Gemini, Grock, Deepseek, and many more models together in one place for just $10 a month. And if you don't know which model's the best, Route LLM reads your request, and automatically selects the right AI for the job.
You can use it to research and write documents, create professional presentations, generate images and videos using leading models, and even humanize your writing so it sounds more like you. But the most powerful feature is its AI agent, which can build full stack apps and complete websites simply by chatting with it. You can even connect Stripe and turn that website into an actual online business without writing the code yourself.
So if you want all your main AI tools in one subscription, visit chatlm.backisai or use the link in the description. So let's get back to the next major development. So here is something from Super Alicia. This is called AGI Roulette and this is an example of the claude models natively coding 3D models in I think it's Blender or Unity or some 3D software and then just asking it to make something. Now, I got to be honest, this is just absolutely hilarious.
Never in my a million years would I think I'm uploading something like this onto the AI grid. But this has been a meme in the uh AI community for the past, you know, week on X. So, if you haven't seen this before, I guess this is where you're seeing this. And I know it's absolutely ridiculous, but the point is is that these AI models can now essentially code 3D games. Now, of course, this is probably the worst example that I'll show you, but what I want to show you guys is just how crazy the models are when it comes in terms of what they're able to do because across a vast range of different examples, they're able to do a ton.
So, for example, this is a short story about an ant and a watermelon, and this is entirely coded with code. So, this image is entirely coded in code, and then Claude has essentially meshed it together and output an MP4 file. And this is the final video. And I got to be honest, it looks pretty cool. Like if I saw a studio in this art style produce something like this, I would say this is impressive. Then of course we had something here.
Again, this is a toy brick building app. You can turn a photo or a description into a buildable model with custom instructions. [music] And this is just something that is really, really cool in terms of the fact that we're starting to see these AI models are now starting to get really really good when it comes to 3D demonstrations of their capabilities. So this is something that is really really under underrated. If you, you know, can find a creative way to use 3D, I think now is a definitely time to do it.
The models are just getting a lot better than people think. And there are a lot of niche use cases that most people simply [music] haven't tried. So, I would say definitely get out there and experiment when it comes to this 3D stuff because from what I've seen, it's definitely gone ahead and surprised me in terms of what it's able to go ahead and build. So, this is something that is super super impressive in my honest regards.
Now once again you can see that there was also this a recreation of the moment Apollo 8's [music] earthrise photograph was t taken using image cues and public data. You can see that essentially Claude is now able to use 3D models. It's able to make video files. It's able to do all of these things very very effectively. And this is something that you know it just doesn't get spoken about because there is no benchmark for video creation.
There is no benchmark for you know 3D manipulation. In fact there probably is but these aren't the benchmarks that are actually the headlines. The only way we realize how good these models are is when we're using them in our day-to-day use cases and we start to see the outputs that are shareworthy and then you start to see these things go viral. So this is another example as well. So Opus 5.5 drew every frame of this animation in JavaScript and it's really really cool because like I said before these animations are completely done by code.
So, it's not like an AI gener. Well, it is AI generated, but it isn't like, I guess you could say, an AI generated video that's simply using those models to, you know, create the final output. This is just pure JavaScript, which is really, really cool. It's this really, really cool paper style. So, I don't know. I I think this stuff is really cool. Now, some other individuals, okay, around the Twitter sphere have been making some very, very impressive stuff.
So, this person, Mangoto, said, "Opus 5.5 is next level. I'm impressed by the 3JS.details. I generated a playable boat scene through Japanese landscapes with dynamic weather, day and night lighting, realistic textures, and 3D characters, and you can view the live sight. So, the water reflections, the physics, the scenery, the architecture is insane. I could see this all day. And this is definitely definitely something that is very, very impressive.
You could see that this game looks like one of those games that I don't want to say AAA game, but I mean, if you add a DLSS filter on this, I think this would look insane. And if you don't know what DLSS 5 is, it's essentially an AI filter that makes games look a little bit more realistic. And so what I'm saying is that like this is a 3JS game and 3JS is not something where you're supposed to, you know, code these kind of complex 3D objects.
But here Claude is doing absolutely effortlessly. Which is why when you start to see the other things it's able to do such as being able to make Minecraft with realistic shaders and Star Wars 3D models, you are just going to be like, "Oh my gosh, this is absolutely insane." So here you can see the Opus 5.5. It makes a realistic Minecraft. And you can see that this on genuinely I thought this is Minecraft. I thought this is one of those tweets where it's like, "Okay, this person is clearly just engagement farming where they're trying to get Twitter impressions." But no, this is actually real cuz I did play a game on Twitter that was like this in the browser and it was exactly like this and I was like, "Wow, this is pretty crazy." And it was literally had the clawed artifact link.
So, the fact that this can now, you know, code these very, very simple games and to be fair, Minecraft, I don't want to say it's a simple game to code. It's not as complex as a AAA 3D 3 3D game, but still this is very very impressive. And if you're thinking, okay, Minecraft is too easy, wait until you see the Spider-Man games. I know we did see a Spider-Man game in the Kimmy release, but this one is dramatically more impressive because it's an entire cityscape.
And only that, you do have the animations and you do have everything as well. So, this is something where this is completely built by Opus. So, it isn't just like, you know, someone imported all the animations, told it to go ahead and build this stuff. This is something where it's it was basically just go ahead and do this. But um yeah, I think this is pretty pretty impressive. No 3D models, just simply coding. And I know it does look weird, but think about where we were a year or two ago in terms of where the models were.
This is why I say this is so impressive. Of course, it's not like a AAA style game. It still looks super super weird and super super awkward, but compared to to where we are a year ago in terms of the models being able to do this, it would have literally been probably a bunch of cubes on a screen with a wire and some moving yellow cubes for cars. So, of course, Jimmy Apples asked Opus 5.5 to make a video of OpenAI solving in Evia Stokes.
He said, "It's a very nice model. I still prefer Astra and 5.6 as my main drivers as I do a lot of research heavy work, but Opus 5.5 is the first time since 4.7 that I want to heavily run Claude. It has great taste." And so, yeah, here you can see that this is something that is very, very impressive. I generally don't even know how it did this. Like, I don't know how it got this background animation. I'm not sure if that's code or how it's been able to do that, but it is very, very visually appealing.
So this is I don't know it's going to be interesting for media because these models are very very good when it comes to doing this kind of content and so you're probably going to see more of this on your feed. Now again we can see that Alex here had early access [music] to Opus 5.5 and he said it blew my expectations out of the water. Here are some things I made and one of the top examples is something called Dark Souls.
So if you never played Dark Souls before, it's like a fantasy 3D RPG. Not really RPG game. It's like one of those games that you know it's kind of you know I mean it's kind of hard to explain cuz I don't really play these style of games but it is just one of the games that looks like this. That's the only way I can describe it. Um, yeah, it's it's really cool. And you can see that this is a vibecoded game. Like this is just purely purely vibecoded.
Like it's not like some complicated game engine running behind the scenes. It's just a vibe coded game. So this is something that I think um for a claude prompt for it to output this, I'm not sure how long it took or how expensive it was. But the fact that the models can do this given enough time and given enough, you know, compute, I think it just goes to show that the models are probably going to be very very capable in the future.
So, um, I mean, think 10 years from now, think 20 years from now, what the models are going to be able to do. That is where your mind needs to be at when we're seeing this rate of progress. For me, this is something that is very, very impressive. I mean, being able to have all of these animations, all of these things. I mean, it's just it's just speechless. Honestly, it genuinely does leave me very speechless. Another example we did have is Ethan Mllock's um test, which he says, "Oopus 5.5 was good in my early tests.
[music] the first non/fable/aststro model to feel like a fableclass model, but still hasn't fully solved the dense language issue of recent plots. And then we had someone being able to make NBA 2K in the browser and they're saying is that ASI made with 3JS and GenX games. So this is something that like I said is going to blow your expectations out of water if you are trying the model. I think is is very pleasant to use.
I've been using it and it does seem like Anthropic are once again back in the hot seat. Hopefully they don't mess up again. There's been so many different things surrounding them. But let me know what you guys think of this model. Are you going to be using this model? Is it better than GBT6 Astra?
The words are the caption track's own and nothing is reworded or re-transcribed. Paragraph breaks are placed between sentences so the text reads as prose.
Free tools for your own script. No signup, no login.
Paste your draft and see where viewers are likely to drop off, with a rewrite for each weak line.
Paste the first 30 seconds of your own draft for a hook score and rewrites.
Check your draft against YouTube's advertiser-friendly guidelines before you record it.
Read this channel's public videos and transcripts, and download a writing brief for it.
| Sentences containing a number | 38 |
Most used terms
Filler phrases
82 in total: like 34 · you know 19 · um 7 · actually 6 · I mean 5 · kind of 5 · basically 3 · literally 2 · uh 1.
A literal whole-word count of the same phrase list the Prepublish browser extension uses, so a phrase inside another word is not counted and a phrase used in its ordinary sense still is. It is a count and not a judgement.