# Did Claude just beat Codex? (Opus 4.8 analysis)

Channel: The Next New Thing
Video: https://www.youtube.com/watch?v=V_g3rlQ0st8
Duration: 20 min
Language: English
Words: 4019
Transcript page: https://viewrankai.com/tools/youtube-transcript/V_g3rlQ0st8

---

[0:00] Opus 4.8 is out. We're going to show you what you as a founder, as a builder, as a leader should care about, and really what doesn't matter to us. We're also going to show you Google Gemini Spark in action. How this is Google's agent right in the cloud ready to work for you. What

[0:13] does it look like inside Open AI's ads? Should you be buying them now? And everything else that's in the news this week right now. Presented by Zapier, the AI automation company. All right, the big news today is that Opus 4.8 is out. You seen it? Oh, yeah. It's It It looks like it's going to be leaps and bounds better than other previous models, certainly better than 4.7, even better than GPT 5.5, and there's even a rumor that GPT 5.6 is right around the corner, possibly this

[0:42] week. So, let's dive in. Yeah, people are saying that this could actually have been called five, Opus five. This is every.2. I like them because they build with AI, and then they also write about their process. They've gone through it. They did what they call their vibe check, where having spent some time with it, they talk about

[1:01] what they liked and what they didn't. And one of the things that they said they liked was that it creates slides. So, what I did was I went to Opus 4.8, and I said, "Turn their article into slides." Here's what it did. We could talk about it. It said, "4.8, four prompts to prove the jump." I hate I

[1:18] hate that that headline. And the first one is the deck tells the story. It even put together a prompt. I don't know. I still find that there's a look here that says that this is the new AI style that everything has. I'm not convinced that every is right that this does better that this does

[1:37] slides better. What do you think? 100% agree. I mean, these slides look exactly like slides that you could generate with Sonnet 4.6 or with Opus 4.7. And to me, this format that we're seeing here, this exact font, and then that little box there with a prompt in it, that to me it has AI written all over it. Like, I feel like that is the M

[1:58] dash of the AI slide generation world. So, to me it's it looks super generic. Okay. All right. The other thing that they did say, and again, this is in my deck. It says it sounds like you, not like like AI. And you did a test, which we're going to show in a moment. They

[2:13] also said it's like a senior engineer. This is where it really shines. It's super strong as a developer. And finally, this is also impressive. 1 million tokens and it remembers it holds the entire code base book length manuscript, weeks of meeting notes in one session. Let me stop sharing my screen so you can share your test of what it was like when you did the when

[2:33] you did the writing test. Yeah. So, what we did is we just did two back-to-back prompts. One with Opus 47 high and one with Opus 48 high. And the prompt that we gave it is my writing style is contained in my brand voice skill. Use that skill and no other skills to draft a three to five sentence tweet in my voice about selling AI services. So, this is the result that we got from Opus 47. It said selling AI services isn't a pitch, it's a translation. Most prospects don't know what AI can actually do for their business. Show them one workflow you'd automate, the tool that runs it, and the

[3:07] hours they'd get back. Do that and they close themselves. Now, to me both of these results you'll see on the the 48 result pretty generic, but considering we gave it no context other than write a tweet. Uh this 47 is not that great, right? Anytime I see a it isn't this, it's this, that to me screams AI. So, right off the bat I'm like this kind of sucks. So, if we look at the response from 48, same exact prompt uh and the output that it gave us is most people selling AI

[3:36] services lead with the tech. Wrong move. Nobody writes a check for I built you an automation. They write checks for I gave you back 10 hours a week. Sell the time you return, not the tools you use. Now, definitely better than the Opus 47 result. Not uh no AI tells in this one, at least you know, according to what I usually look for. Again, generic, but we didn't give it much to work with. So, just from this very limited basic test, Opus 4A definitely a better writer than

[4:04] 4.7. Okay, fair. I use it for writing a lot. I'm glad to hear that this is actually an area where we're going to see some strong use cases. Let's take a look at what what else is being said about it from people who have tried it. This is Wade Foster from the founder of Zapier, of course my sponsor. He's had some time. I really admire that he gets all this extra access to it. He's saying, "Look, uh according to their to Zapier's benchmarks, it is a front and center. The scorecard here is their performance based on their benchmarks. They're saying Opus 4A beats Gemini 3 Flash. It beats um

[4:39] uh GPT-5.5. This was all expected. Here's what he has said about it, though. He says, um "The biggest unlock for agentic workflows, refusals where the model declines to act, dropped from 20% to 4%. 4.7 would often decline sensitive tasks outright instead of reacting or partially completing them. 4.8 keeps it going." That's really impressive. That's

[5:03] important. Overall, big step forward. Um big news. Anything else to say about that? No, I think we pretty much nailed it. Again, step in the right direction, certainly better than previous models, but not a quantum leap in capability like a lot of people sometimes expect. So, we're inching a little bit forward with every model release, and I think

[5:24] the next one will be that much better. Yeah, and we're actually going to see in a moment that maybe it doesn't really matter if you're using the latest and greatest all the time for everything. But, why don't we go on to the next piece of news, which is Gemini Spark. People have had a chance to play with

[5:37] Gemini Spark recently. And what this I kind of think of this as like Claude Co-work for Gemini. I really like this uh Rob the AI guy's uh explanation of how it worked here. to go through bl101.com and make a spreadsheet of all of their different products and the price. And now, this will actually do is this will be able to go through this website. This will be able to actually access your Google Sheets, and it will be able to actually keep track

[6:11] of this and update this. And guess what? We can actually get this to do this. Now, let's take a look at the result of that. And the beauty is look at this. This has literally gone through and scraped so far hundreds of different products, given us the price, given up the audience and category, sizes available.

[6:29] And this is pretty insane. The thing that I like about it is it's accessing web pages, it's interacting with Google Sheets. The negative that he said was the problem is that it can't connect to a lot of different apps. And so, it's still fairly limited, but it takes action. And I can show you a couple of other things about it. What do

[6:47] you think of Gemini Spark so far? So, I mean, it's just Google, exactly like you said. It's Google's version of Claude Co-work. It's all the all the model the big frontier model labs are going to come out with their agentic co-pilot tools. You see Microsoft Co-pilot, Claude Co-work, Gemini Spark, uh Perplexity Computer. I think Chat GPT has an agentic tool, but nobody really uses it. So, I mean, this is again, this is par for the course at this point. Uh it really comes down to which of these co-pilots are going to be better at computer use, and which ones have more out-of-the-box connectors. Because the average everyday user's not going to set

[7:25] up an MCP server to connect to a tool that doesn't have a native connector, even though it's literally copy-paste a single link. Um so, yeah, step in the right direction for the cuz because Google has that distribution already built in. And here's what I like about them versus Co-work. The thing that I can't stand about Co-work is that my computer needs to be up and running in order for it to

[7:45] work. I hate that. I really hate that. And don't get me wrong, I keep my computer running all the time. But, you know, if there's a glitch somewhere, if I happen to turn it off, if for whatever reason it's not working, I don't want things to not be available. And this is very cloud-based. And I zoomed in a moment earlier just to

[7:59] show that it's right there in Gemini. You just switch over to Spark, and I like that one of the things that people keep doing with it is they get their daily briefs. Here, let me let me have him talk about the daily brief that he's got. So, now, the century what this does is this is connected to your calendar, this connected to your email, this connected to your Google Drive. And every single morning, this actually goes through. We can see it did it this morning at 1:48 a.m. And this actually goes through a bunch of things. We have what's top of mind, hey FYI, looking ahead, and then

[8:30] this goes through a bunch of different things that I need to do. For example, I need to refine a YouTube video strategy for a specific thing. And what I could actually do is click on this, and now this will automatically, using my Drive, my Calendar, my Gmail, wherever it actually got that information from, actually begin to do work for me. Now, if you want to turn this on, what you need to do is come into your settings right here. You're going to click on personalize intelligent Bef- before we go on, it's not available for everyone yet. So, if you don't see it yet in your Gemini account, you need

[9:00] to upgrade to Ultra or wait for it to to make its way into your account. But, this is how he sets it up. In order to have this be as most powerful as possible, that memory is turned on, the daily brief is turned on, that if you come into connected apps that you at least have your workspace turned on, but ideally, you want all of these different things turned on. In addition to that, if we actually come back out over here, you also want to make sure that instructions for Gemini, that you have anything in there that's specific to your daily brief and the things that might be created by this

[9:32] daily brief. But, that was just the first update. The second update that we want to check out is now if we come over into Spark right here, we're going to see Gemini Spark. And again, co-work. It's interesting that every time somebody has this new type of technology, the first thing they do is they create a um

[9:49] Daily brief. The daily brief. The daily brief. I was like, that seems to be like the Yeah, the go-to like workflow test for any agentic tools. Like what kind of daily brief can I create? I think you're you're right with that. All right, on this computer I don't have my logged in uh account to the information, but essentially what the information has said is that OpenAI is now going after smaller advertisers with ChatGPT ads. That means that SMBs are now going to be able to go in. We found one SMB who's gone in and looked at it. I really like how his whole channel, Henry's, is focused on ChatGPT

[10:22] ads. This is what the ads look like. configure the ads. And what I really like about what ChatGPT has done here, they've got this little preview on the right-hand side. Now, as you can see, we really don't have a lot of space here, so our copywriting really is going to be on point. So, you want to make sure you've got concise, real clear headlines and descriptions, as well as amazing creative that's really going to catch

[10:44] the attention. What do you think so far of ChatGPT ads, Corey? So, I think this is going to open up a ton of opportunities for SMBs. I mean, think about when Meta ads first launched. I mean, this is like the early days of Meta ads, but times 10 or times 100. It's going to be really cheap in the beginning, and then everybody's going to flood in, and then it's going to get really expensive and a lot more difficult to compete. But, as a side note, I mean, this guy uh positioning his entire YouTube channel around ChatGPT ads, I think is genius. Like that's the first thing I

[11:16] noticed. He's only got, you know, just under 7K subs, but I mean, that's the type of channel where you just go super niche, and he could be the the guy for ChatGPT ads. So, hats off to Henry. Yeah. Um he did bring up that there are some issues there, but let me show you what

[11:33] you can do for targeting with it. most of you are going to be wanting to create the campaigns manually. Now, when you create the campaign, you have the ability to add a campaign name and then select between the objective that you want. Now, for now there is either reach or clicks, which means you're either going to be paying by CPM or cost per click. And you can actually see in the future they are potentially going to be releasing cost per conversions. And if you actually head over to the Open AI developer center, you can see that they've actually got all of this measurement information here. They've

[12:07] got a All right, all the things that we're expecting they're making their way over and from what I'm seeing from my founder friends, they want to be first into this and so they're willing to put up with a few restrictions in order to get there early before others figure it out. It will be interesting to see who who scales this quickly. Like this is still so new. I'm sure it's going to do really well, but it's just a matter of who can spend the fastest in a row as that makes

[12:33] sense. Grok wants to be the agent of choice for for coders. What do you think of Grok Build? What is it? So, it's XAI's new agentic coding tool. Uh the video that we'll show here in a second, I like what I like about this video is it puts it to the test and I can kind of give my synopsis uh maybe after we watch the video. This guy pushes it to the limit, spawns 200 sub agents. So, let's let's see what that

[13:02] looks like. 200 sub agents and do a complete performance. Oh, woah, look at this. This one woah, woah. They're all crashing. They all just crashed. Oh my gosh, he couldn't take the sub agents. Wait, too many files. Wait, what just happened? It just crashed. thread SES panicked at crates codegen XAI Grok Shell SRC session.

[13:28] All right, we'll leave it there. He he had found issues like that a lot. Let's get to your analysis of it. Yeah, so the quick and dirty of grok build is based on sentiment from folks who are in the weeds with it. It you know, it's a good model. It's not up to par with the frontier models. You get rate limited a lot quicker. So this guy ran it for an hour and already hit 8% of his monthly usage limit on the

[13:56] $300 a month plan. And you know, it's good, but it's not great. So it's still early on. They have a lot of catching up to do. At this point, Claude coder codex is still going to be the gold standard, but I I'm not one to bet against Elon. So I see this getting a lot better a lot

[14:11] quicker. Good point. Let's get into this other news. I really like what Matthew Berman said about using cursor. I don't think that really it's destroying Anthropic's Claude as it's shown here, but check this out. This is He makes a good point. The point is imagine how much more you're going to be able to get done for the budget. I think In fact, here, let me highlight what he's saying here. He's saying, "Over here, you see Opus 47. It is really powerful, but it's also expensive." He's looking at $11 per task here. Versus um GPT-5, it's almost as powerful as Opus 47, but considerably cheaper. We're looking at a

[14:52] little bit over four bucks. And Cursor's composer 2.5 is not as powerful as either of those, but it's significantly cheaper. And this is what he's getting really excited about, Matthew is. And we're talking about a difference in power of two percentage points from composer 2.5 to you know, Opus 47 on max, right? So it's like you're getting slightly slightly slightly less performance, but for quite literally 10% of the cost. So, unless you're just working on a project that requires absolute you know, top-of-the-line inference and ability, why would you use Opus 4.7 Max? You're going to be spending 10 times the money for marginally better performance compared to Composer 2.5 in this

[15:41] scenario. It's a real eye-opener. I I like what he's saying. I've been reading on Reddit that people are are starting to really like Composer 2. It's there. It's less expensive. Most of us, I don't think need the most expensive model, but we we pay for it anyway. He compared it to Gemini. Here, let's listen to this. Just a a moment more about the comparison to Gemini 3.5 Flash. Now, keep in mind that Composer 2.5 and Gemini 3.5 Flash came out basically at the same time. And I I think there were a lot of people let down by Google I/O's announcements. Gemini 3.5 Pro did not come out. They said they needed another

[16:21] month to bake it. Gemini 3.5 Flash seems like a lot of people didn't like it or or didn't really appreciate it. Flash is a general model. It is not specific to coding, although they talked about it a lot as related to coding because they were hyping up Antigravity so much, and it was the main model in Antigravity. Yeah, they kind of put themselves in a position where they were going to be compared as a coding model. So, another thing I asked Sundar All right. And when he compares it to what Cursor has, it falls flat. All right. Final news story. This is going to be super exciting for a lot of

[16:57] people, and some people are just not going to care, but here it is. Robinhood is now allowing your agent to trade, and it actually will give your agent a credit card. This, I think, is really interesting, exciting news if you're if you're looking to play around with automated trading. I think this is the future. I mean, this is where everything's going. You've got things like agent mail. You've seen 50 different startups that give your agent a credit card or a bank account. So, you know, agentic trading via the biggest platform out there like Robinhood seemed to be the logical next step. So, interesting to see how these agents

[17:35] actually perform. I mean, there's going to be entire VC-backed startups dedicated to agentic trading agents. Um but yeah, the infrastructure's there, which is the step one. So, we'll we'll kind of see how quickly the results follow. And and also how quickly, you know, the majority of trading becomes agentic versus humans. I know a lot of it's done by algorithm already, but what when is the majority going to shift to agents versus just regular

[18:02] algorithms? Already as I look at the top GitHub repos of the week, I'm seeing more and more uh tools to help your agent trade better automatically. This is a an interview that Robinhood CEO Vlad did with A16Z about do you get an advantage if you are trading with agents? And here's what he said.

[18:22] We'll do this in a few seconds of it. things we do for fun or the things that we do as a time jobs. Do you think people actually can invest for a living? And if everyone's using advanced agents to trade, the edge in a stock market is someone knowing something that someone else doesn't. If all the agents know exactly what to do, how do you think

[18:38] agents can actually meaningfully trade? You know, everyone thinks that this is going to be all great and my agent's going to be trading for me. Um and and I think what happens with all technology is the early adopters do have an easier time. So, you know, you're going to get some early wins possibly if you're doing some kind of agentic trading because not a lot of people are doing it. At some point, if everyone's doing it, it becomes almost like if you're not doing it, you're at a disadvantage. Mhm. It's kind of like uh when you're the first accountant that used the spreadsheet in

[19:10] the '80s. It's like, "Oh man, I can get all of my work done for my business in 1 hour a day. Yeah, so temporarily it's going to be a huge advantage for people who jump on it. Overall in the future, it's just going to be another another thing and we'll all have it is what he's saying. Yeah, he's just basically saying that it's a leg up for now and very soon it's going to be table stakes. So, makes

[19:34] sense. All right, we're still experimenting with this so I'd love some feedback from people who are listening about how it's going, Corey. This almost didn't happen today. The internet in my whole part of Austin went out. I had to rush out to a WeWork, which is where I am. You can kind of tell when it's a WeWork when you get this

[19:49] wallpaper, don't you think? Yeah, oh it looks good though. You we made it happen and luckily we did. We we hit record at 90 minutes after Opus 48 came out. So, we had to scramble scramble to to grab that headline. Yeah, ended up working out. All right, thanks. And now that we're done, I got another video for you to watch right here. The one that Corey and I did where he showed you how if you're using Cloud Skills, if you're using Cloud Co-work and you're not using this setup process that he created, you're missing out on all the power of it and people love it. It's one of the top

[20:19] videos here on the channel by um by measure of how many people ask to use the thing that they saw and you'll see it and you'll get to use it immediately. Go check that out.

---

About this transcript
Read from YouTube's own caption track and laid out by ViewRank AI (https://viewrankai.com).
ViewRank AI finds the videos already beating a creator's own average on Instagram, TikTok and YouTube Shorts, transcribes them from the audio itself in more than 60 languages, and turns what worked into new ideas and scripts.
Free transcript tools, no account needed: https://viewrankai.com/tools
How to read any video this way: https://viewrankai.com/llms.txt