Opus 5.5 vs. GPT-6 Astra Showdown

Eric Siu (0:00)

Claude Opus 5.5 versus GPT-6 Astra, which one actually wins, especially when it comes to doing actual work? So what you see on my screen right here is a comparison done by Artificial Analysis. And you can see the different cost per intelligence and intelligence index. And you can see for this one up here, this graph over here, which is the intelligence index versus cost per intelligence index, you can see that GPT-6 Astra, let's just go to medium over here, seems to be a little lower when it comes to cost per task. And you also can see the intelligence index that Claude Opus 5.5 is higher, and that medium is higher over here. It's actually a little less cheaper. So what I did for this experiment was we focused on medium because medium is a little more practical for real life. So it's the same exact prompt and we just let it go. We just let it interpret things.

And we just want to see how well it figures things out on its own. And keep in mind, this is using medium. Not only that, we gotta look at costs as well before I start to show you some of these examples. $4 per million input tokens and $20 per million output tokens. And it also lists at $10 and then $50, right? So it's significantly higher. And we're gonna first start with the packaging part. So right now with the packaging piece, so this is this one over here. So we can see what the packaging looks like. And these are the thumbnails. That's what it is. And so this is a thumbnail saying, this is what you should fix first, right? So I'm gonna show you the three thumbnails that I came up with. Here's another thumbnail on Backlog to Brief. So this one over here, Show the Source Content Draft Approval Blocking. So what do I think about these? I don't think these are great, if I'm being honest. First, it does this whole AI written thing, where it kind of makes you feel like a dummy, right? Like three promises, three ways in. Like, what does that even mean? Fix this first. What does that mean? Backlog to Brief. What does that even mean? Show the source. What does that even mean?

This doesn't quite make sense to me, so I'm curious to take a look at what Opus put together, right? So let's take a look at what Opus put together, and then we can decide on a winner here. So if we look at this one, this is the tab where Opus did some work. So when the click over here, so it says four AEO mistakes. So is this the best thumbnail ever? No, but at least I'm in the picture, not that I need to be in every picture, but at least I can understand where it's trying to go.

Call should move to drafts over here. So Content OS will pull from your best content, whether it's internal meeting sales calls and then a drafts room. So it's at least explaining the logic to me a little bit.

And I guess it did a revision. So here's the original version and then it did a revision, but they look the same to me. But what am I going to say about this? I'm going to say this is a lot more usable. Again, they both had the same exact prompt and it seems that Opus 5.5 had a little more horsepower here. So you can see it even explains its reasoning as well over here. I'm like, that's cool. So in this case, I have to give the first round to Opus 5.5. Keep in mind, Opus 5.5 Medium and then GPT-6 Astra Medium. That's what we're looking at. So round two, let's go ahead and look at round two. So round two again, I'm going to explain it real quick. So round two is an ad tear down. So we're back on ChatGPT over here.

And you can see that the directive here is to inspect the supply ad and then rebuild around an offer the business can actually fulfill and then competitor performance is unknown. So what is this over here? So let's take a look at the ads. So ad number one is this, Signal Foundry.

So the platform for every content job. So this is just like a fictional ad.

This is planning, drafting, refreshes, reporting one workflow, many possibilities. So we have a workflow for this, which does AEO and SEO.

I'm like, well, I don't quite understand that. And then you have three different test creatives. Your backlog isn't a priority list. So again, this is just AI slop, if I'm being honest with you. So my interpretation so far, based off these first two examples, is it looks like the horsepower from a thinking standpoint is a little bit lower when it comes to GPT-6 Astra Medium. So I'm going to move over to the second one, and we can take a look at the second one being Claude. So let's take a look at that one. And you can see for Claude, here are the ads that it drew out. These are the teardowns, right? And so you can see this one, 60 deals in flight, zero time to type, okay? Talk your way through admin, works in Salesforce, Gmail, and Panda Dog. Bad prompts equals bad results from ChatGPD, okay? Speak rich. So this is like, I think this is pulling from Wispr Flow. That's what it is, right? So this one over here, meaning notes to get the details right. I think this one actually came up with good ad concepts for Wispr Flow, and I think it just decided to pick something versus just, I don't know, for ChatGPD, it kind of just didn't really come up with a concept to go after, and I guess maybe it's pulling from one of the workflows that we have. So I look at this, I'm like, okay, look, there's more bad prompts equal bad results from ChatGPD. AI skips your brand. Here's one that got you cited, okay? So basically, maybe it's looking at Wispr Flow over here. It's looking at what's working and it's saying, hey, here's what you should go after over here for Single Grain. AI skips your brand. Here's one that got you cited, 400 plus more AI citations for Ellis Building Products. So yeah, that's actually pretty good. I understand what it's doing now. It's actually saying, hey, based on these Wispr Flow ads, these are what we think you should go after. And then here's three options that you should go for, which I think are right here. So solid SEO, missing an AI search. Okay, 400 plus citations. Read the case study.