Gemini 3 vs. Claude Opus 4.5 vs. GPT-5.1 Codex: Which AI model is the best designer?

I put three cutting-edge AI models to the test in a head-to-head design competition. Using the exact same prompt, I challenged Google’s Gemini 3, Anthropic’s Opus 4.5, and OpenAI’s Codex 5.1 to redesign my blog page, evaluating them on visual design quality, user experience improvements, and SEO optimization capabilities. One model produced a beautiful, polished, production-ready redesign. One was fine. And one completely whiffed. If you’re trying to figure out where each model fits in your workflow—design, planning, back-end, or something else—this episode will save you a lot of trial and error. What you’ll learn: How each AI model approaches the same design challenge differentlyWhy planning capabilities dramatically impact design qualityThe specific visual and functional improvements each model madeWhich model excels at front-end design versus back-end functionalityHow to strategically choose the right AI model for different parts of your workflowThe importance of model-switching based on specific use cases— Blog design: https://www.chatprd.ai/blog — Brought to you by: Lovable https://lovable.dev/—Build apps by simply chatting with AI — Where to find Claire Vo: ChatPRD: https://www.chatprd.ai/ Website: https://clairevo.com/ LinkedIn: https://www.linkedin.com/in/clairevo/ X: https://x.com/clairevo — In this episode, we cover: (00:00) Introduction to the AI design challenge (01:25) The question: Which model is the better designer? (03:08) The prompt used for all three models (04:10) Gemini 3 Pro’s approach and results (06:00) Opus 4.5’s approach and results (10:54) Codex 5.1’s approach and disappointing results (14:51) Comparing the three designs side by side (16:03) Analyzing the change logs and SEO improvements from each model (22:43) Final verdict (23:00) Conclusion and next steps — Tools referenced: • Gemini 3 Pro: https://deepmind.google/models/gemini/pro/ • Anthropic Opus 4.5: https://www.anthropic.com/news/claude-opus-4-5 • OpenAI Codex 5.1: https://platform.openai.com/docs/models/gpt-5.1-codex • Cursor: https://cursor.com/ — Production and marketing by https://penname.co/. For inquiries about sponsoring the podcast, email [email protected].

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:04Claire Vo?

  Welcome back to How I AI. I'm Clara Vo, product leader and AI obsessive, here on a mission to help you build better with these new tools. Today, I have a really fun mini-episode where I'm going to answer the question on everyone's mind. Which of these new models is actually the best designer? I'm going to take a page on my site that I don't think is particularly well-designed and have Gemini 3, Opus 4.5, and Codex 5.1 duke it out and see which one can redesign my page better, one-shot. Let's get to it. This episode is brought to you by Lovable. If you've ever had an idea for an app but didn't know where to start,

• 0:43Claire Vo?

  Lovable is for you. Lovable lets you build working apps and websites by simply chatting with AI. Then you can customize it, add automations, and deploy it to a live domain. It's perfect for marketers spinning up tools, product managers prototyping new ideas, or founders launching their next business. Unlike no-code tools, Lovable isn't about static pages. It builds full apps with real functionality, and it's fast. What used to take weeks, months, or even years, you can now do over the weekend. So if you've been sitting on an idea, now's the time to bring it to life. Get started for free at lovable.dev. That's lovable.dev. If you've been paying attention in the last couple of weeks,

• 1:28Claire Vo?

  it seems like every single model provider has released a brand-new coding model. And what I heard the most from people is, sure, they're fast, and sure, they're great, and sure, they're beating benchmarks, but they are all really good at design. If you've been on X or social media, you've probably seen these beautifully-designed landing pages, apps, and user experience components.

• 1:51Claire Vo?

  Generated using Gemini 3, or Opus 4.5, or even Codex 5.1. And I thought, let's put these side by side and actually see which one's better at redesigning an existing page. I think it's easy to one-shot something and make it look beautiful, especially if you're a great prompter and know exactly what to say as a designer. But if you have an existing site and you want to make it better, who's your trusted design engineer? Which of these models is really going to do the trick? And I'm going to show you what I think today in a couple of minutes on which of these models is the better designer or redesigner of a page that I don't think is really great.

• 2:32Claire Vo?

  So this is the ChatPRD blog. It is not very good. I don't think this is a very beautiful site. It's not my favorite. I think it could be a lot better. And it could be a lot better from a functional perspective, but it can also be a lot better from a design perspective. And, you know, if I had a team, which I have a little small one, but if I had a team that was not AI, I might send this to a designer and say, hey, we just launched this. Early on, it's not great. Can you redesign it? And so I wanted to test that flow with some of the new models that have come out that have said that they are better designers than previous versions. And so I fired up Cursor and I did a model by model comparison of redesigns.

• 3:16Claire Vo?

  And I used the exact same prompt, exact same input code, and we're just going to see which one we think is the better designer. So I'm going to show you my prompt here in Cursor. It was pretty straightforward. It was this, redesign the blog page. So I just showed it the directory of where our blog page is to improve both the visual appeal and user experience. So sort of both like, will it look nicer and will it be functionally a little easier to use? And then I added a functional component to it, which was add best practices for SEO and navigation. And then I did that for three different models. I did it for Gemini 3 Pro. I did it for Opus 4.5 from Anthropic and I did it from GPT-5.1 Codex.

• 4:01Claire Vo?

  These are all recently released models that have been said to be their best in class models from OpenAI, from Anthropic and from Google. And so we're going to see exactly what it did. And I started with Gemini 3 Pro. The reason why I started with Gemini 3 Pro is I've heard over and over and over again, what a great designer Gemini 3 Pro is. And I really wanted to see what it did. And so you can see here, it thought quite a bit about visual design, user experience, SEO navigation. It looked at the code and it started executing. So it started writing some code. And we're going to switch over and see exactly what it generated.

• 4:42Claire Vo?

  So it generated this. This was the before, if you recall, it's very, very boring, not very good. And in the after, it generated a nice hero image of the most recent blog post. So there's now this like highlighted blog post at the top and then these cards at the bottom. And a couple of improvements I see here, there's some tagging here, there's some date of releases. There's this nice hover effect that zooms in on our featured images when you zoom in. Haven't done anything regarding pagination, which is a current functionality that doesn't really take into account whether or not we have featured images and making that look good. So there's some things there that could be improved, but I think overall it's pretty good.

• 5:27Claire Vo?

  One thing that I noticed that it did that I did not love is that there's this tag at the very top of the page and it's just a little too tight with the rest of the navigation. So one of my reflections here is, you know, it doesn't have like the full visual context of the page, but it did a pretty nice job and it was very fast. But I have to say, despite Gemini 3's reputation for being the best designer, it was actually not my favorite. So I ran the exact same query in Cursor with Opus 4.5. So

• 6:08Claire Vo?

  If you look up here, redesigned the blog to improve both the visual appeal and UX and add best practices for SEO and navigation. Now, the difference that I thought was really interesting when using Gemini 3 versus Opus 4.5 is Opus 4.5 actually triggered a to-do list inside Cursor. So it did a tool call to, to create a to-do list and it gave a step-by-step flow it was going to follow. So Gemini 3 sort of did that chain of thought reasoning and then just you'll load code. Opus 4.5 created four to-dos. So the to-dos were redesign the blog listing page,

• 6:47Claire Vo?

  improve the blog layout, enhance the post display, and add comprehensive SEO, structured data, canonical URLs, and meta tags. And so it was very precise, step-by-step on what it was going to do in terms of implementing. And so I think the planning capabilities of Opus 4.5 are certainly better. I think Anthropic has really differentiated themselves as experts in coding models. You know, if I wanted to get the best outcome here, I probably should have done this in Claude code because I think there's some optimizations they've done there recently as well. But I thought it was really interesting that the output of a planned implementation was much

• 7:28Claire Vo?

  better than the output of a straight shot, one-shot implementation. And so you can see, it went step-by-step and actually checked off those changes and then provided me a summary of changes. And I'm going to switch and show you exactly what that looked like because I was actually impressed by, by the design. So this is what we got from Opus 4.5, which I think, spoiler alert, from all the models, was the most beautifully designed blog page that I got, and also honestly, the most functional from an SEO perspective. And so what you can see

• 8:10Claire Vo?

  that Opus 4.5 did here is it pulled some images. We have a repository of beautiful background images and featured images that we use throughout the ChatPRD website. It actually pulled and looked for assets that it could bring in that would look nice. These rings are some... Design elements that we use commonly. And so it pulled in some interesting assets. If you recall Gemini 3 just had a gradient background. Opus 4.5 actually added some imagery in the background. Very similar concept in terms of the layout. So you see again, a featured article. That is the most recent blog post. Again, three column cards with the zoom in trick. So I guess people like it.

• 8:55Claire Vo?

  But if you look at this, a couple of nice design tweaks that Opus 4.5 added. When you hover, not only does the image zoom in, but it gives you this nice little call to action here, this little arrow. I think it is so cute. Just does that nice little touch hover treatment on the anchor link for the blog post. Again, tags are in, and then it did a little bit more on the SEO side. And I will wrap back around to the SEO changes that each of them made. But if you see here, not only do you have the author, which is me, it's Claire. You have the date, which we also saw in the Gemini 3 option, but it also has an estimated time reading and a link.

• 9:36Claire Vo?

  And so I just think the quality of the design here went probably 20 or 30% further than the Gemini 3 model went. And it's those nice edge touches that I feel like AI can add into any design that just makes it so much nicer to work with. And I was really impressed with Opus 4.5 in terms of the quality of the detail orientation. Now let's go down. You know, one of the things that it did is it handled no images a little smarter than Gemini 3 did. So if you recall Gemini 3 kind of collapsed these cards here, did not put placeholder images in. Here with Opus 4.5, it saw that we were missing images for some of our blog posts

• 10:19Claire Vo?

  and put a little placeholder with a nice little book icon here, which I think is lovely. It makes these cards just look a lot nicer and is really well designed.

• 10:32Claire Vo?

  So overall, I think that Opus 4.5 did an excellent job out of the box of redesigning a page and not only redesigning the page, but really thinking about the functional components of it. And I think a lot of that goes to its planning mode and its ability to call tools and then do some of these implementations step by step. Now let's get to the last model that I tested, which was Codex 5.1 Pro. So again, same prompt here. Redesign the blog to improve the visual appeal and UX and add best practices

• 11:09Claire Vo?

  from SEO. Add a GPT 5.1 Codex, the leading coding model from OpenAI. Again, Codex, like Opus 4.5, thought and generated to-dos. The to-dos were a little less granular than the one from Opus. So if you look at Opus, the to-dos were redesign the blog listing page with specifics about how it was going to redesign, improve the blog layout, enhance a specific component, and then add SEO. The plans for 5.1 Codex were a little bit more general. They were investigate current layout, redesign, apply SEO. So I think the planning

• 11:52Claire Vo?

  was just not as thoughtful from a design perspective as the planning was from Opus 4.5. And then if we actually look at the design, "OpenAI, you know I love you," some of my favorite models, but it did not do well on this redesign. And so you can see a couple things that it didn't do well right out the gate. One, it gave me AI slop purple gradient. Like, we do not need any more Purple to blue gradients in AI designs. We need to get them out of here. And so just the fact that we got AI purple is an immediate disappointment.

• 12:34Claire Vo?

  The other thing, and this may be a me problem, but I think we have a white wordmark and a better logo to use here. And you can tell here, just the image it selected is not nice on top of a colored background. Now, I do think that the headline and copy from the, um, from the blog is really nice. Stories, playbooks and experiments from the team. Um, so it gives a little bit more context. So this was the model that did the best copywriting perhaps, but overall the design was not very good. And then again, it did featured post here. This is the image from our most recent blog post, but there's no context. There's no call to action. It doesn't link to anywhere.

• 13:18Claire Vo?

  And so I'm just really unsure what it was expecting users to experience. Now it's repeated here, um, the featured blog. So again, I think these, I think these models really like, I guess there aren't that many fancy things in blog design and that you all have to have a featured image and then a three row, um, layout for your blog post. So it did do the featured image here, but the problem is it added a bunch of these links that don't really, I don't understand how they work. They only do the featured image in each of these categories. The jumping's kind of weird. And then if you look at it at browse the library, it doesn't even show the blog posts that exist in our overall, um, library.

• 14:02Claire Vo?

  And so it's both not pretty. It's purple and it doesn't work. And so I was really surprised because I've had pretty good experience with, um, GPT five and five one in functional sort of backend work, but the front end work, it just really struggled. And I will tell you, this is not a complicated app. This is a basic, you know, blog layout with a basic CMS on the background. It is nothing that is technically complicated. And so what I would say from a GPT Codex five one perspective is it's not going to be the designer on your team. It has another role to play on your team. And I have found plenty of places for this model to be really, really useful.

• 14:47Claire Vo?

  But design is not one of them.

• 14:52Claire Vo?

  And so I would say, just looking back, Opus 4.5, absolutely my favorite from a front-end design perspective. Gemini 3, very serviceable, could probably benefit from some planning and implementation. And then Codex 5.1 is just not your front-end girl. So we got to get something else in the front-end. And what I like about testing these models on a specific use case like this, where it's repeated, is you can start to understand which model goes at what part of your workflow. I'm a real believer in model switching. I know everybody has their personal preferences, but I think there are great models for writing. I think there are great models for design. I think there are great models for image gen. There's great models for planning and strategic thought. And I think there's great models for backend coding.

• 15:37Claire Vo?

  And not all of these models are created equal. They're all exceptional. I mean, think about the work that they can do on behalf of teams. But they're not all the same. And I think as you test them out, looking at similar use cases over models, and making a decision about where you're going to place a model on a team, is a really important skill to have as you're developing your AI fluency, as a designer, as a product manager, and as an engineer. Now, I want to go through the functional side of things before we wrap up this little mini app, which is going to be a true mini app, hopefully under 20 minutes, which is summarizing the changes you made. So I asked each of the models to summarize the changes they made in terms of design changes, SEO changes,

• 16:22Claire Vo?

  and just what did it do? And so, you know, I like this as a workflow, as you're working with coding agents, especially if you're running a lot of them and you're not paying attention to them. Asking it to summarize the changes it made, so you can compare them, are really useful. And so if you look at Gemini 3, it made a new hero section, which we know. It made feature post layout, which we know. Glass morphism card. Thank you. Thank you, Apple, for giving us glass morphism. I think we could live without it, but it's at least a standard, likable design style. So it has scaling images, deepening shadows, improved typography, related articles, and visual breadcrumbs. Now, let's look at this because one of the things I did not check

• 17:07Claire Vo?

  is if these models actually changed how the blog post themselves showed up. So let's click into that and see if there were layout changes that were made to the actual blog posts. And there were. Okay, so Gemini 3 did make some changes to the actual blog posts, and it said it added related articles. Okay, so that's a little bonus, is it went beyond just the blog homepage, and it added some SEO functionality into the blog post itself. Now, let's read the rest of the changes. From an SEO perspective, good, JSON LD, great SEO schema that we definitely want, breadcrumbs, which we definitely want, semantic HTML, which is really helpful,

• 17:48Claire Vo?

  especially in blog, and then related articles and metadata. So lots of very helpful, I think, high-quality SEO changes to my blog post from Gemini 3 Pro. So I'm going to give it a little bit more credit, and then it went a little further than I initially analyzed, and actually went to the blog pages itself. But let's check that against my favorite, which was Opus 4.5. So I'll look at Opus 4.5. What changes did it make? Now, see, these changes are extensive. So again, I think that planning mode really allowed it to make very specific changes across a variety of components. So it made feature posts in three-column card grid, which we know, the little arrow slide-in that I noted, reading time badges, category

• 18:33Claire Vo?

  pills, breadcrumbs, which we like, and graceful empty states. So these are all things that I identified when I was visually scanning the design that I thought was really nice. The blog layout had this nice rings pattern, improving spacing, and then the post display has more information. So let's actually see what it did on the posts, if anything. So let's click through.

• 19:02Claire Vo?

  So it made, again, very similar changes to what we saw in Gemini 3. So again, like, don't redesign everything. If you are doing something like a blog, you're gonna get best practices. So it brought in that metadata in terms of author, date, and reading time. Let's see if it added those anchor links. It did not add any related links. So it maybe didn't do as great of a job on the SEO on the individual article pages, but it did a really nice job redesigning the call to action at the bottom of our blog post, which is something that I don't think Gemini did. So it added, I'm sorry to say, it is, again, AI purple slop. So we gotta say, no more purple, especially ChatPurdy is so pink. It should know it.

• 19:46Claire Vo?

  It should see pink everywhere in my repo. It should do this. But other than the purple, I think this is a really nice call to action for a newsletter subscribe. There's a subtle gradient in here. There's a drop shadow, this little kind of avatar call out next to how many product managers are subscribed. Actually, there's like 90,000 product managers subscribe, so we gotta update the content there. I think this is a really nice little component. And this is another thing I've noticed about these new coding models is we're all getting wowed by these beautiful page designs and app designs. What is really impressive is you give it like a small little component, a little widget, and have it redesign, it looks so much nicer. So that's what it did from a design perspective.

• 20:31Claire Vo?

  Let's see from an SEO perspective. So metadata, again, open graph, structured data. Let's see if it did JSON LD. It didn't specifically call out JSON LD. So I'm gonna have to check to see if it did that. That's one important part of our SEO roadmap at ChatPurdy we've been working on. So surprised not to see it. But again, maybe you put Opus 4.5 in the designer mode, and you put some of these other models in your like SEO engineer mode, and then another model in your sort of like backend engineering mode. So maybe we just have figured out where each of these models need to live. Now let's do our last one and look at Codex 5.1. What were the changes it made? Now, this is the shortest summary.

• 21:15Claire Vo?

  And again, this is the one that did the worst job at this, I will say, also.

• 21:22Claire Vo?

  GPT-5 models love a bullet point. If you see a bullet point, this is a 5 or 5.1 response here. And so I asked it to categorize the changes you made. Again, use the exact same prompt. It gave me five bullet points. Very lazy. So hero panel, category chips, featured article layout, and then SEO changes did metadata and embedded a schema.org. So it did the JSON LD block. So that's good. So again, we weren't really impressed with the Codex 5, GPT-5.1 Codex model on design, and actually not that impressed on the details in terms of user experience and SEO. So I think maybe this guy belongs in the backend. I probably could have prompted it better. But again, the

• 22:07Claire Vo?

  point of this mini episode is to show, if we have a basic prompt, the same way I would speak to a colleague, that I don't have time to tell exactly how to make better. I'm hoping they can research and understand how to make a page better. I would just say, hey, our blog is not good. We need to prove the SEO. We need to prove the UX, and it needs to be prettier. Can you just take care of it? What it would do. And that's how I like to think about these models, is how do they respond to these natural requests you would make in the day to day of your work, and then compare how they do on the outset. So to recap for everyone, we did a, we started with

• 22:47Claire Vo?

  a existing layout. It was not pretty. It was not functional. It was not good. We gave a three line prompt to redesign it for UX, visual appeal, and SEO. And then we compared three models. We compared Google's Gemini 3, Anthropic's Opus 4.5, and OpenAI's GPT-5.1 Codex. And the winner was for sure on the design side, Anthropic's Opus 4.5 model, both from a design perspective, as well as a usability and SEO perspective. And it went further than even my prompt requested.

• 23:30Claire Vo?

  The hypothesis here is both it is better trained on high quality front end design, as well as its detailed planning allows it to do a much better job on the details and implementation than these other models that do more shallow planning or no planning at all, as we saw in the Gemini 3 And so we just got a better outcome. I love my new blog design. I am very excited about this. If we just take a step back, it is incredible that in less than 20 minutes, we were able to generate not one, not two, but three alternative designs for an existing website. We were able to get massive upgrades on the functionality of it, especially some

• 24:15Claire Vo?

  technical SEO stuff. And I was able to pick the one I like. Imagine asking your teammate to design you three different options, give you three different plans for SEO, and then tell you which, you know, have to go back and forth on which one you like better. I think this is an awesome flow. I loved it so much. I'm actually just gonna go ahead and ship this today. So we'll put it in the show notes so you can see exactly what happened. And that is my takedown of which of the new models from November 2025 is the best designer. And I think the winner is Opus 4.5. Thank you so much for joining this mini episode of How I AI. I cannot wait to share more tips and tricks and hands-on experience with AI, and I will see you soon.

• 24:59Claire Vo?

  Thanks so much for watching. If you enjoyed the show, please like and subscribe here on YouTube, or even better, leave us a comment with your thoughts. You can also find this podcast on Apple Podcasts, Spotify, or your favorite podcast app. Please consider leaving us a rating and review, which will help others find the show. You can see all our episodes and learn more about the show at howiaipod.com. See you next time.