Ep 504: Has Anthropic’s Claude lost its edge? What happened & can Claude recover?

Topics Covered in This Episode:

1. Anthropic's Claude Losing Market Edge
2. Comparison of AI Innovators: Anthropic vs. Google
3. Claude 3.7 and Industry Relevance
4. OpenAI and Google AI Advancements
5. Enterprise Hesitation with Anthropic's Claude
6. Benchmark Performance: Anthropic vs. Competition
7. Claude User Experience and Rate Limits
8. Future Prospects for Anthropic and Claude

Keywords:

Anthropic, Claude, Claude 3.7, Claude's edge, AI model, Large Language Model, AI frontier labs, Google, OpenAI, Microsoft, Apple, Synthetic data, Differential privacy, Tariff policies, NVIDIA, U.S. AI manufacturing, Taiwan Semiconductor, Foxconn, Wistron, Digital twins, Advanced robotics, GPT 4.1, Million token context window, API developers, Context processing, Claude's competitiveness, Internet access, Third-party integrations, Rate limits, Claude artifacts, Claude projects, Computer use, Power users, AI safety, Claude Max, Rate limit sessions, Cloud recovery, AI coding index, Coding performance, Price comparison, Benchmark performance, Intelligence index, Human preference, Elo score, Web traffic, Market competition, Enterprise challenges.

Podcast Transcript

Jordan Wilson [00:00:16]:
I would say for the better part of two years, the large language model race was three teams. You had Anthropic, OpenAI, and Google racing for the lead and going back and forth jab for jab, as the best AI model maker in the land. Obviously, you know, Microsoft's in there, but they're more of a system that uses other technology, but when it came to actual AI frontier labs, it's always been a three team race. I don't know if it's like that anymore. I think right now, open AI in Google are so far ahead of everyone else. And I'm left wondering what happened to anthropic. What happened to Claude? Is it still a top tier large language model or has Claude completely lost its edge? And can they ever, catch up with Google and open AI? All right. We're gonna be talking about that and a lot more on everyday AI.

Jordan Wilson [00:01:31]:
What's going on y'all? My name is Jordan Wilson, and I'm the host of everyday AI. This thing, it's yours. It's your daily live stream podcast and free daily newsletter helping us all not just learn AI, but how we can leverage it to grow our careers because you can try to keep up with AI news and developments and new large language model updates. You can try to keep up, but just hearing about them, reading about them doesn't do anything. You need to leverage it, and that is what our website is all about, youreverydayai.com. So there, we recap each and every day's podcast episodes. Sometimes I have guests on. Sometimes it's just myself.

Jordan Wilson [00:02:08]:
So we bring you exclusive insights every single day. We're actually the only AI newsletter that does that as well as we keep you up with everything else happening in the world of AI. So you can be the smartest person in your company or your department when it comes to generative AI. Alright. Let's actually do that and go over a quick recap of what's happening in AI news for April 15. So Apple is responding to criticism over its AI performance, particularly in areas like notification summaries, with a timely pivot, toward synthetic data and differential privacy. So, yeah, Apple, kind of responding, according to reports by focusing a little bit more on synthetic data. Right? So the company generates, according to the report, the company is now generating synthetic data to emulate user information without using real content, enabling private testing on datas of users who opt into device analytics.

Jordan Wilson [00:03:10]:
So this approach ensures accuracy while safeguarding privacy. So, yeah, Apple obviously has had a super, super slow rollout. And by super slow rollout, they're years behind everyone else. And their Apple intelligence, let's just say it has not been well received. So, some new reports and information are showing Apple's kind of new or updated approach, by using synthetic data, kind of tying it to those who sign up for, kind of this device analytics. So by pulling devices with synthetic data comparisons, Apple is hoping to enhance, its Apple intelligence with better email summaries and other functions, signaling a broader commitment to addressing user concerns and advancing its AI capabilities responsibly. Alright. Our next piece of AI news, NVIDIA has committed $500,000,000,000 to US AI manufacturing amid changing tariff policies here in The US.

Jordan Wilson [00:04:12]:
So NVIDIA announced plans to invest up to 500 billions in, $500,000,000,000 in AI infrastructure manufacturing within The US over the next four years, marking a significant shift in its supply chain strategy to meet surging demand for AI chips and supercomputers. So the move coincides with US president Trump's ever changing tariff policies, which initially impose steep levies on imports from Taiwan and China, but recently exempted chips in other tech products, easing concerns for companies like NVIDIA and Apple that rely heavily on overseas productions. So NVIDIA will partner with Taiwan Superconductor, in, Taiwan Semiconductor in Arizona for chip production, and with Foxconn and Wistron in wiz in Texas for supercomputer manufacturing, aiming to achieve mass production at these facilities within twelve to fifteen months. So by using digital twins of factories and advanced robotics for automation, NVIDIA hopes and plans to streamline operations and enhance efficiency in its US based facilities, demonstrating how AI technology can transform the manufacturing process. So, yeah, if you're wondering like, okay. What the heck does this matter? Well, so many big companies and all the AI systems that we use like JAD GPT, Google, Microsoft, everyone else. They're struggling to keep up with demand. Right? So, essentially, everyone's looking for more compute.

Jordan Wilson [00:05:39]:
This is a pretty big move from NVIDIA to bring more, kind of AI power, to The US. And then last but definitely not least, OpenAI has launched a new family of models with the GPT 4.1 series. Probably the big headliners there is it's now has a million token contents window, but right now, at least, it is only available for a p on the APIs, end. So only for developers right now. So OpenAI has launched its new family of models, g p t 4.1, as a major upgrade to its previous models offering advancements in context processing reliability and cost efficiency. But like I said, you're not gonna find it. If you go to chatgpt.com, it's not there. At least right now, OpenAI did not announce any plans for it to live, on the front end inside chatgpt and is only, available for developers on the back end.

Jordan Wilson [00:06:38]:
But let's talk a little bit about the model because I have some pretty pretty impressive specs here. So, GPT 4.1 introduces a 1,000,000 token context window, far surpassing GPT four o's previous tops on the API end, which was a 28,000. So that's big. So, you you know, Claude and Gemini and others were really beating OpenAI historically in context window, right, but not anymore. So pretty pretty big news there. And then unlike previous models, integrated into chat g p t, like I said, g p t 4.1 is exclusively available through OpenAI's API, making it a tool tailored for developers rather than general use. The performance is pretty impressive across, coding, instruction following, and complex reasoning tasks. What's also important is OpenAI has said some of those improvements have also been rolled out kind of under the hood, to its, GPT four o model.

Jordan Wilson [00:07:42]:
I would assume it was the late March update that there wasn't a lot of updates about, and there are now three new varieties. So there is GPT 4.1, kind of the full version, GPT 4.1 midi, which is, more affordable and compact. And then GPT 4.1 nano. Yeah. The first time, you know, OpenAI has gone at nano, and that is their smallest, fastest, and cheapest model. Yeah. If it's it's as if it's not hard enough to already understand these models, now we have, two variety of small ones. Yeah.

Jordan Wilson [00:08:15]:
If you thought mini was small, no. Now apparently, mini is medium and nano is small. And then some sad news for some old school, you know, if you like some of these old models, OpenAI is planning to phase out older models like the o g GPG four, by April 30. And then also somewhat surprisingly, OpenAI announced they'd be phasing out GPT 4.5 preview by July 14 to focus on the more efficient 4.1 lineup. Also, this release coincides with a delay in GPT five's launch now expected in a few months as OpenAI navigates some integration challenges. And yeah. So FYI, obviously, OpenAI has changed, course a couple of times. They essentially said, hey.

Jordan Wilson [00:09:05]:
We're gonna stop, releasing non reasoning, models, and GPT five is gonna be more of a of a hierarchy or a system. So they said, yeah. We're not gonna be releasing a lot of new models before GPT five, and here we are. So, alright. Let's get into it. A lot more on those stories, on our website at youreverydayai.com. What's up livestream crew? Yeah. If you listen on the podcast, come join us sometime on the livestream.

Jordan Wilson [00:09:34]:
You you know, when I have guests, we take questions. Sometimes I ask you all things. So, thanks to everyone for joining in. George from YouTube, big bogey, saying GPT 4.1 is a coding powerhouse. Yeah. It is already early benchmarks. Trade printing here from YouTube. Thanks for joining us.

Jordan Wilson [00:09:53]:
On the LinkedIn machine, Kimberly, Dennis, Allison, thank you all for tuning in. But let's just get straight into it. Has Anthropic's Claude lost its edge? It's Tuesday, y'all. I'm gonna take a sip of my coffee and let me know. Should I crank this up? It's been a while since I really brought it on a hot take Tuesday. I'm a little tired, but livestream audience, if you could, leave me an emoji or two. Should I be one fire emoji? Should I be kinda nice? Two fire emoji. Should I bring the heat? Or three fire emojis burn, baby, burn.

Jordan Wilson [00:10:32]:
I mean, I don't know. One thing and let me tell you this. I tell you all the truth. I do. Period. Right? As an example, if you would have asked me eighteen to twenty months ago, Hey, Jordan. What are your thoughts on on Google Gemini? I'd say, don't use it. Ask me today, Google Gemini is the king of the hill.

Jordan Wilson [00:10:57]:
Right? I I do think it is Google and OpenAI now going, jab for jab, but I tell you the truth. Right? So I'm not I'm not gonna hold back if you all want, a little bit a little bit of fire. Alright? Rolando here is saying to crank it up. Fred. Alright. Fred. Fred. Thank you.

Jordan Wilson [00:11:15]:
Fred's Fred's like, alright, Jordan. Be nice today. He wants me to be kinda nice. Allison, here throwing in some, some dynamite. That's dangerous. Alright. Alright. We'll see.

Jordan Wilson [00:11:31]:
We'll see. You know? I don't wanna I don't wanna offend anyone because let me say this, let me say this. Claude is still one of the most impressive pieces of AI technology ever created. All right. Period. So I don't want to overlook that. All right. But what I've found is I've been using Claude less and less.

Jordan Wilson [00:12:02]:
I would say probably nine months ago, Claude probably accounted for about 25, of my usage. It's probably down to about 5% now. I'm finding it hard, to find actual use cases for Claude. And I'm talking about on the front end. Y'all all right. So I'm not talking about on the back end. I know that, Claude three five has historically been, you you know, one of the most used models if you look on, like, OpenRouter. I know that, Claude three seven is still popular for developers, although not it's not the most popular anymore.

Jordan Wilson [00:12:41]:
It's not the most popular anymore with, Gemini two point o flash and Gemini two point five pro. It's really not, but this has been a long time coming. So back in September, back in September, if you want to go listen to this, what episode was this here? Three fifty one. All right. So I told y'all back in September, '3 reasons businesses shouldn't use Anthropics Claude yet. And this was after like a year, right? This was a year of, of, of me being hesitant. So what a lot of people don't know, people are like, okay. Jordan's just some random guy that, you know, jumps on a podcast and talks about AI.

Jordan Wilson [00:13:27]:
Well, yes. On the surface. Right? On the other end, I do a lot of things that you all don't see on this show. Consult big companies, companies with tens of thousands of employees. I work with research organizations. They reach out to me, big ones, big name ones, and they're like, hey, Jordan. Can you help us better understand generative AI? So it's it's it's much more than, you know, this this, little podcast. Although thank you all for listening, you know, and making Everyday AI a a top 10 tech podcast in The US.

Jordan Wilson [00:13:58]:
But I'm talking with a lot of businesses, a lot of things that you don't hear, and it's not just me. Big enterprise companies have always been hesitant to use Claude at scale. All right. And it's been a long time coming. I even said three big reasons. This was back in September that I said Claude was in trouble and enterprises shouldn't be using it yet. Number one, there is no enterprise access. So I'm talking about on the front end.

Jordan Wilson [00:14:31]:
Right? So keep that in mind, everyday AI it's for largely non technical people. Right? And I'm talking about logging on to, you know, claw.ai, or I'm talking about logging on to gemini.google.com, chat g p t Com. Right? Using this on the front end with your team. One thing I'm a huge advocate for, if you listen to the show, is having your AIOS. Right? Your AI operating system. Your team needs one. Right? In addition to whatever your company may be doing on the back end, you need a front end AI operating system where you and your team collaborate to get work done. No internet access.

Jordan Wilson [00:15:11]:
They, Claude went the first two years without having internet access. They just, rolled out internet access about a month ago. Okay. Very limited third party integrations. Alright? Google technically on the front end has not a lot of third party integrations, but because they're Google. Right? Because they have, you know, anything that could be a third party integration, they essentially have in house. Right? Google has, like, a trillion of their own products. Right? Extremely limited third party integrations.

Jordan Wilson [00:15:46]:
It's improved since September since I had this show, episode three fifty one. And then I said extremely restrictive testing tiers. So both free and paid. I'd say the one biggest thing that's been knocking Claude off, in in from real business adoption is you can't even go and test it. You like, if you have a paid even a paid plan of Claude. Right? And and and you're like, alright. You know, let's go test this. Let's see if this is right for our business.

Jordan Wilson [00:16:15]:
You know, you're paying the $25 a month or whatever. There's been, and I'm not exaggerating, hundreds of cases because I use large language models. I mean, it it varies. I don't know. Anywhere from four to twelve hours. Recently, it's been a lot of twelve hour days using large language models. Right? It's so easy on a paid plan to hit your rate limits on Claude. I kid you not within ten minutes.

Jordan Wilson [00:16:39]:
It's happened to me hundreds of times where I will hit on a paid plan the rate limit within ten minutes. Yes. I'm generally working in multiple tabs if I'm in Claude. I'm working with long context windows. Yes. I can't tell you the last time I've hit chat GBT limits. Right? Doesn't happen. Gemini doesn't happen.

Jordan Wilson [00:17:05]:
Claude has been extremely restrictive. And I think that was a major misstep early on. How do you expect aside from, you know, appealing to your, to your core audience, which we'll talk about because I think they're losing, space there. Right? You know, coding, development, software engineering, etcetera. Right? How are you going to appeal to the average business owner, to the average enterprise use case when, you know, a company pays and maybe they get a team plan? I think those those rates are about double, but still you can't even use the thing when you pay for it. It is extremely restrictive. All right. And another reason why I think that Claude has lost his edge.

Jordan Wilson [00:17:52]:
It's no longer innovating right in the early part of 2024, even midway through the year. I still think Claude was an innovator, Right? They came out with artifacts, which when it came out extremely impressive. So if you don't know Claude artifacts, it's actually kind of hidden. You have to like enable it and then you have to make a call to it. Right? But it's it's it's still and right now, let me be honest, because I still said Claude is still one of the most impressive pieces of AI technology. There's still great use cases. Right? Even though I'm trying to, you you know, y'all wanted the flame emojis, I'm not gonna totally poo poo on Claude. There's still some use cases.

Jordan Wilson [00:18:33]:
I said, maybe now it's five to 10% of, you know, my use. But the only thing I use Claude for right now is using 3.7 thinking on artifacts. That's it. Nothing else because everything else. Claude is not a top model anymore. In many cases, it's not even a top five or a top 10 model, which sounds crazy to say because nine months ago, they were that, that, that tier one, right? If we go back to our, our ranking tiers, right? Like S a, b, c, right? They were us. They've fallen how the mighty have fallen. But that's that's all I really use it for.

Jordan Wilson [00:19:18]:
But Claude and Anthropic were innovators, you know, early on. So the artifacts so, you know, that's something that can render code, in natural language. You know, you can have it build you a business dashboard, you know, games, whatever. Right? And you can run it in the browser. And then guess what? ChatGPT and Gemini said, alright. Yeah. Let's do this as well. So they came in with Canvas.

Jordan Wilson [00:19:43]:
Alright? Say similarly, Claude was an innovator with projects. Anthropic innovated with projects. Right? A good way to, you know, organize your chats, a good way to, leave custom instructions and, project knowledge. Right? ChatGPT follows suit. Computer use. Right? Anthropic was innovating. Although back in October, when that came out, it was extremely clunky. Extremely clunky.

Jordan Wilson [00:20:13]:
Right? You know, one of the easier ways, to to run computer use was, you know, you had to download Docker. You had to go to GitHub. You you know, work with their repo, which is fine, but for non technical people, not that good. In the rent limits, I did a live show where going over Claude's computer use, again, very hard to use with the rate limits. So I think Claude is no longer innovating. Now I think they're clone chasing. Whereas before others were copying their innovation, now they're copying other people's innovation. So, yeah.

Jordan Wilson [00:20:50]:
Like, now a lot of the things that you see and that are gonna be rolling out, like, as an example, according to, some online sleuths right now, Claude is testing voice mode and all these things. Right? They're just now seemingly cloning features that were popular six months ago, a year ago. And one of the reasons I think is Anthropic dropped the ball. Right? Back in September, when I gave those three, those those three reasons, those three three, different scenarios on why I thought enterprise companies and why I told countless enterprise companies don't use cloud for those three reasons. They didn't entrust those. Those were not secrets. It was no secret that you couldn't it was so hard to literally use cloud on the front end. They knew that.

Jordan Wilson [00:21:43]:
Right? Their their, you know, team is interacting with people online on on Twitter. You know, everyone's been complaining about work, rate limits and, you know, Claude's team has been saying, oh, we're working on it for years. It's too late. It's too late. One of the reasons why. Right? I don't know this, but we've heard stories that as an example, OpenAI is losing money. Right? Even CEO, Zay Maultman said on their new pro $200 a month subscription that they were losing money even though it's been extremely popular. So I don't know.

Jordan Wilson [00:22:17]:
This is my hunch, but my hunch has been, Claude has been maybe more profitable at least by, percentages than maybe their main and closest competitor to chat GPT. But at what cost? Because I don't think they're growing their user base. I don't think about it. You know, I think sometimes, you know, if you are an avid listener to this show, if you're, you know, an AI nerd like myself, we I mean, we all live in an echo chamber as well. Right? Outside of our little echo chamber, no one knows about Claude. Right? But they could have, they could have a year ago if Anthropic would have listened, to its customer base a little more closely and continue to innovate and improve the product, improve usability, I don't think we'd be having the same conversation today. Always have receipts, y'all. Always have receipts.

Jordan Wilson [00:23:24]:
Alright. So on my screen, this is January 2025. Web traffic. Alright. No one uses Claude. Comparatively, no one uses Claude. They don't. I know I'm gonna catch them flat catch some flat for that and, be like, Jordan, you're, you know, a a chat GPT fanboy or, you know, jumping on the Gemini bandwagon.

Jordan Wilson [00:23:53]:
No. I'm not. I've been using Claude since the day it came out. I've enjoyed certain features. I use all, you know, I I I've used dozens of LLMs and, like I said, hours every single day. Claude's not good anymore. It's not. I got more stats.

Jordan Wilson [00:24:11]:
I got more receipts. Don't worry y'all. You said you wanted some flame emojis. Right? So let's look at total visits in January 2025 web visits, chat gpt.com, 3 point 9 billion with a b. Yeah. With a b. Claude, seventy six million. Gemini, two hundred and sixty seven million.

Jordan Wilson [00:24:48]:
DeepSeek, two hundred and seventy seven million. So, you know, essentially, Gemini and DeepSeek are are right right there with each other in terms of people visiting the front end. Perplexity, 99,000,000. Y'all, ChatGPT. Let me let me do some quick, some quick napkin math here. ChatGPT has more than 10 times the users of Claude, Gemini, DeepSeq, and Perplexity combined. Is my math right there? 500. Alright.

Jordan Wilson [00:25:20]:
Almost. Sorry. My math my Natkin math was was a little wrong there. Alright. So we have, that's that's about 500,000,000, six hundred million. Alright. So, about five times. So Chad GbT has five times more users than Claude, Gemini, DeepSeek, and Perplexity combined.

Jordan Wilson [00:25:38]:
And Claude is in at least according, to, kind of online demographic or online, website information, which is pretty accurate. Right? I've been using these, different SEO tools for, ten plus years. They're very accurate. No one's using Claude. Hot take. Ready? It's been less than two months since Claude released its latest model in Claude three point seven. Claude is on it 3.7 and it already feels antiquated. So they announced it February 24, Claude '3 point '7.

Jordan Wilson [00:26:16]:
And let me just call this one out. Right? They're they made a big deal of of Claude being, you know, the world's, you know, first hybrid model. Right? So, you know, when you think of old school transformers and then you think of these, you know, quote unquote new school, models that think in reason under the hood. I don't know. To me, that seemed like a marketing gimmick from anthropic. Right? Why? Well, you have to actually if you want to use that extra thinking, right, you have to go in and you have to click the button. So is it actually a hybrid model? I don't know. I'd say not.

Jordan Wilson [00:26:58]:
So now I think you also have anthropic falling down this trap that Google fell in, in late twenty twenty three where they're getting caught up in the marketing and not listening to their users in shipping new capable powerful models. But claw three seven Sonic feels antiquated because since that time, we've already had multiple updates from OpenAI. We've had multiple updates from Google. We've even had multiple updates, from models that I'd say never use like DeepSeek. Right? If you care about your privacy, don't use it. Unless you're, you you know, downloading it and fine tuning it locally. Right? But don't don't use DeepSeek on the web or their API if you care about your data. If you are a business, don't do it, especially in The US.

Jordan Wilson [00:27:42]:
Right? But, anyways, how are we at the point now where a model that is not even two months old feels antiquated? That's where we're at. And I don't know if Anthropic can keep up. Like I said, they're very innovative to begin with. They're great researchers. You know, obviously, I think they are a world leader, in terms of of AI safety, in terms of, ethics, right, all of those things. But in terms of, like, okay, are they just gonna be more of a of a of a research arm that kind of drops AI models, or are they trying to actually dominate? Are they actually trying to be relevant? Are they actually trying to be one of the top large language model makers in the world? I don't know. I was personally very underwhelmed with Claude three point seven SONNET, their newest model. Even the thinking variation when you have to toggle it on.

Jordan Wilson [00:28:39]:
I know a lot of people I was reading, you know, online, you know, a lot of people are using, are using it inside, like like George here on YouTube says, you know, he says Claude seems lazy in Wind Surf and Cursor, but it is not when you use it in an app. Yeah. So I know a lot of people yes. Claude, up until, you know, a week ago when, Google said, oh, wait, Claude. You are no longer relevant because we're dropping Gemini two point five pro, which wipes wipes all the way the competitive advantages that Claude three five or Claude three seven SONNET had. Right? Google just said, yeah. We're we're we're gonna knock you off, this pet this pedestal. You're not gonna compete.

Jordan Wilson [00:29:23]:
Google straight up wiped them, which is interesting. Right? Because Google is, you know, has invested, you know, but they're still technically competitors in some regards as well. Let me tell you what I mean. And here's my hot take. I think there's a lot of Twitter talk and hipster hype when it comes to Claude three seven or Claude three five, but I care about business utility. Anthropics lost its edge there. I care about benchmarks. I care about real human usage.

Jordan Wilson [00:30:02]:
Claude's not competing there anymore. And like I said, I think one of the biggest things that's happened in that I would not want to be working at anthropic right now, Gemini two point five pro and 2.5 flash. I don't know if I'm being honest, unless anthropic has been sitting on a world changing model, I don't know how Anthropic is going to compete against Gemini two point five pro and Gemini two point five flash. Good luck. I know, you know, a lot of people have said, oh, well, there's still, you know, Claude three seven Opus. Right? You know, InfraPic Claude had kind of these three tiers of models. They have their small haiku, their medium, Sonnet, and their big one Opus, and they haven't updated Opus in a very long time. So I was like, oh, you know, cloud three seven Opus or, you know, cloud four point o will I I don't know.

Jordan Wilson [00:30:52]:
I don't know because there's also rumors even though, Gemini 2.5 Pro just, went generally available, like, ten days ago. There's already rumors that Google has a much better and more capable model that they're already testing on the l m chatbot arena. I don't know how Gemini is going to compete against Google. Alright. Got receipts as always. Y'all I have receipts. Yes. Similar web Dennis.

Jordan Wilson [00:31:19]:
Thank you for asking. That's where that data was from. All right. Let me know y'all. Why stream audience? Am I am I wrong on this? But let's get quickly to the receipts. Alright? I'm not gonna make you wait an hour for this one. I'm gonna go through quickly because the proof is in the pudding y'all. The writing's on the wall.

Jordan Wilson [00:31:38]:
Alright. So let's look at artificial, analysis. So a great, third party unbiased website, right, that does benchmarks. Because one of the thing is when companies put out their benchmarks, they cherry pick. There's dozens of different benchmarks. So, of course, you know, when when, you know, these AI labs put out their models, they choose, okay, out of these 50 benchmarks, here's the eight that we're gonna put on our website because we look great on this. Right? So I always look at Elo scores. We're gonna talk about that in a minute here from Ella Marina and look at third party benchmarks as well.

Jordan Wilson [00:32:14]:
So intelligence, this is from the artificial analysis intelligence index. Gemini two point five pro in the lead. Second, o three Mini High from OpenAI. Then you have, the two variations of deep seek, and then you have the new version of GPT 4.1. Y'all, I'd account. Cod free seven is number eight in terms of intelligence on this third party benchmark. Let's keep going because you're like, okay. What about humans? Humans probably prefer it.

Jordan Wilson [00:32:48]:
Okay. So Elo scores. Let's talk about that. That's head to head. You put in a prompt on LM arena on the chatbot arena. You get two outputs. You don't know who they are. You say this one's better.

Jordan Wilson [00:32:57]:
Alright? There's been millions of votes. Guess what? Total Elo score. Claude is not a top 10 model. That's when I well, like, I know it sounds crazy to say, but you have to ask the question even if you ask it rhetorically. Is Claude no longer a state of the art model. I don't know y'all in so many benchmarks in so many now elo categories, overall elo, they're not a top 10 model Gemma three, which is a small language model from Google has a higher Elo score. Then Claude three point seven. Let me say that again.

Jordan Wilson [00:33:46]:
A small language model, not a large language model. Humans prefer the outputs across millions of votes compared to quad 3.7. Google has, let's count it, one, two, three, four, five models. Five different models that humans prefer over quad three seven Sonnet. I don't know. Also, I don't know. It's my hot take very hot take when I said, hey, anthropic has lost its place atop. Right? Gemini two point five pro, higher.

Jordan Wilson [00:34:24]:
Let's see. We have Gemini two point o flash thinking, higher. Gemini two point o pro experimental, higher. Gemini two point o flash, higher. And then there's small language model, Gemma three. My gosh. Alright. But you might be saying, alright, Jordan.

Jordan Wilson [00:34:37]:
Well, people use Claude for certain reasons, right? They use it for creative writing. Claude's great at that. They use it for coding and software development. Claude's great at that. That's an old narrative. Literally, that's an old narrative. Right? Especially the creative writing thing. I think essentially, right, you know, a bunch of stuff went viral online, like, maybe a year and a half ago, about how bad chat g p t and, and and Gemini were at writing content, and Claude was just so much better.

Jordan Wilson [00:35:12]:
Alright. Well, let's look at those two things. Let's first look at creative writing. Okay. Oh, where's Anthropic? Oh, the bottom of the list. Again, not top 10 elo in creative writing. That's what I'm saying. I think right now it's a lot of Twitter talk and hipster hype.

Jordan Wilson [00:35:33]:
Right? Oh, it's cool to like Claude. Right? It's like, oh, you you know, I see you wearing that name brand, chat g b t. Oh, I see you with that mainstream Google Gemini. I'm over here prompting with that Claude, man. No. Why? Why? Not a top 10 model when it comes to creative writing, which everyone thought it was amazing at it was a year and a half ago. Don't lie with me. Y'all this is, this is millions of people have voted this blindly.

Jordan Wilson [00:36:08]:
Guess what else? Not a top five in coding either. It's not. Claude three point seven SONNET with thinking is not a top five model for coding. Guess what is? Guess what's at the top? OpenAI is o one, they're o one preview, they're o one mini, Gemini two point o flash. And I do, I I do believe once, Gemini 2.5 pro, is on here and gets enough votes, it'll be up there as well, but not a top five model in terms of coding. So what do you want? What do you want? I don't understand. Why are people still using Anthropic? Like I said, maybe you have one or two use cases that you're happy with it. Right? If I'm being honest, the only thing I use it, like I said, Claude used to be maybe 20% of my usage.

Jordan Wilson [00:37:02]:
I'm a heavy large language model use, user. Like I said, maybe 5% now. I'm only using it because there's certain things in artifacts, that that Claude does better, than Google's Canvas and OpenAI's Canvas. But it's always like I'm doing it all at the same time anyways. I'm running the same thing in all three of them, and sometimes I'm like, okay. Yeah. Anthropics is a little bit better here. Alright.

Jordan Wilson [00:37:27]:
So maybe you're like, oh, it's it's it's fast. It's affordable. It's not fast. It's not affordable. It's not you know, when you look at speed, and this is from artificial, artificial analysis, Gemini two point o Flash and Gemini 2.5 Pro are the fastest models followed by GPT four o and o three Mini from OpenAI. Again, claw not in the top five when it comes to speed. Alright. Which is output tokens per second.

Jordan Wilson [00:37:52]:
So it's not fast. Alright. And that is the non thinking model, by the way. Alright. It's terrible on price. It's terrible on price. Right? Which I I I still don't understand why people are are are so, deep sea drunk. Like, deep sea is not cheap anymore.

Jordan Wilson [00:38:14]:
Right? It's not. When it first came out, it's like, oh, yeah. This is cheaper. Okay. Well, Gemini two point o Flash is wiping, the floor with everyone when it comes to price. Llama's new, llama four scout, g b d four o mini. Right? There's just so many, faster, better, cheaper models than quad. So I don't it's definitely lost its edge.

Jordan Wilson [00:38:36]:
Right? I think there's one more thing I wanted to to pull up here. Okay. It's coming up in in in a slide here. Because this is also telling. So looking at the intelligence versus price. So it's not like you're getting a good bargain either if you're using Claude on the back end. Right? You're not. You're not.

Jordan Wilson [00:38:51]:
So on the front end, humans aren't preferring it. On the back end, you're not necessarily getting what you pay for. Again, this is intelligence versus price, so there's a little quadrant here. So you wanna be on the upper the upper left because that means it is, cheaper and smarter. Claude is on the right side, and, Claude three point seven SONNET is actually on the bottom, right? All right. Not necessarily fast or affordable. And here we go. Everyone's like, oh, it's the best coding model.

Jordan Wilson [00:39:25]:
Guess what? It's not artificial, analysis. They're coding index. This one is very interesting. Claude three seven, the thinking model. Ready? The thinking model is in fifth place. Guess what's ahead of it? The new model that was just released from OpenAI, GPT four one. But guess what? Y'all this is the mini version, the mini version of open AI's new model. Not only is it a non thinking model, right? Because normally if you use these thinking models, these reasoners, they code much better.

Jordan Wilson [00:40:07]:
Right? Especially when you're working, with very complex tasks, in long token, long contact windows. So not only is this GPT four one model, it's not a thinking model and it performs better on artificial analysis coding index, but it is the mini version. It is the mini version. So I don't know y'all if you're still using quad three seven SONNET. Let me know why. Let me know why. I'm very curious. Like I said, I know a lot of people on the software engineering side, on the, development side.

Jordan Wilson [00:40:46]:
They love it. Right? Using it with cursor, using it with windsurf, using it, you know, inside all these different IDEs. I also don't understand why on that. Now with Gemini 2.5 pro, with Gemini two point o flash, and now these new models from OpenAI that they just announced, I I don't I don't understand it. I honestly don't understand how anthropic has gone in in Claude has gone from that top tier, right, state of the art world leading model to kind of irrelevant. So a lot of people are like, Oh, well, you know, Claude just released a new plan, Jordan. You're you're you're really harping on them for, you know, these rate limits. You can just pay more and use it way more.

Jordan Wilson [00:41:35]:
Okay. Well, why? If it's not a top 10 model right? Yeah. Claude just came out with their, Claude Max. Right? So you get, higher limits, you know, if you're paying a hundred dollars a month or $200 a month, which let me just call this out, you know, because people are like, okay, Jordan, this this solves. Well, you don't get anything more powerful for that 100 or $200 a month. You don't get more features. Right? So when OpenAI, as an example, announced their $200 pro plan, at the time, that was the only way you could access Sora. That's still the only way that you can access o one pro, and then you get unlimited everything.

Jordan Wilson [00:42:11]:
Unlimited. This is not limits. Or sorry. This is not unlimited. You can still go in on the front end, and pay a hundred dollars or $200 a month. You don't get new features. You don't get new models that are exclusive to that max plan. You just get slightly better limits.

Jordan Wilson [00:42:29]:
But here's a concerning one y'all. This one's kind of concerning. Ready? This is from Anthropic's, website talking about their new plan. Ready? Talking about their message limit on the new max plan. Your message limit will reset every five hours. We call these five hour segments a session and they start with your first message to Claude. Please note that if you exceed 50 sessions per month, we may limit your access to Claude. Each session includes any messages sent within five hours from the first initiated chat.

Jordan Wilson [00:43:13]:
So we expect it to be fairly generous for our users. My gosh. I don't know. How tone deaf is this y'all? Come on. So let's just say in theory, let's say you're a very regimented person. Alright? Like I am. So this is my I can't even use Claude on the current paid plan, but even if I pay a hundred or $200 a month. So let's say I use Claude in the morning before my show to help plan it.

Jordan Wilson [00:43:38]:
Alright? So let's say 6AM, and then I use it at noon, midday. Alright? And then in the evening, you you know, I use it again. So let's just say I just do a couple of props. Couple of props a day. I do it at, you you know, 6AM. I do it at, you know, noon, and then I do it at 6PM. Six, noon, six. Right? Couple of prompts a day.

Jordan Wilson [00:44:00]:
Paying a hundred or 200 or $200 a month. In that scenario, even if I'm only doing a couple of prompts, right? Paying a hundred, dollars 2 hundred a month, I might get cut off from my pricey 100, dollars 2 hundred a month plan. That's what they're saying. 50 sessions a month. So if I do that, if I use Claude three times a day that are more than five hours spaced apart, I could, in theory, in three weeks, get shut off. And I, you know, I might not be able to use their paid plan, for the last week of the month. In theory, that's what it's saying here. How tone deaf is that? I don't understand.

Jordan Wilson [00:44:40]:
If I'm being honest, when I saw that, I'm like, come on, anthropic. You you you have I don't know. How many billions of dollars, have you gotten from from Amazon? I lost track, dollars 6,000,000,000 or something. This is why people aren't using your service. Humans don't prefer it. Benchmarks don't prefer it. And for those people that are actually still finding utility in our power users, you're slapping them in the face. Get real.

Jordan Wilson [00:45:10]:
All right. Hot take let's end it here. Can Claude recover? I honestly don't think so. I don't think so. Here's again, this is just reading reading reports. You can't knock Anthropic for putting safety first. You can't. They put out world leading research.

Jordan Wilson [00:45:47]:
I do think when it comes to, you know, safe AI, they are a leader in that. But no one's paying you for your research. You're not competing to be the the the best frontier AI lab with the best research, with the best safety. This is a race. This is the wild west. Right? That's what it is. There's no rules when it comes to AI. Anthropic is playing I'd say the wrong game.

Jordan Wilson [00:46:28]:
They've alienated their power users. They've stopped innovating. And I think that has caused them to now face an almost insurmountable challenge. Right? Let's just say, as an example, Claude had their four point o model ready, and they probably have had it ready for a while. When you see these new drops from OpenAI, right? They're 4.1 models, the smaller versions when it comes price per performance, amazing. Same thing with Google Gemini 2.5. I don't think if I'm being honest, right, where nine to fifteen months ago, I'm like, yep. It's gonna be a three team race.

Jordan Wilson [00:47:18]:
It's not anymore. Yes. You have to pay attention to open source. You have to pay attention to Chinese models, but most enterprise companies here in The US aren't gonna touch, many open source models for different reasons. And they're not gonna touch, Chinese models for obvious reasons, data security, data privacy, and not sending all your business IP straight to China from a US perspective. Infropic was primed to compete in this three team race. They were primed to be a a a leader, but now they're a second tier company. They are.

Jordan Wilson [00:47:57]:
That might be harsh. You wanted my honest take? That's not just me. Is that my personal usage? Sure. Is that my personal experience? Yes. But I showed you the receipts. Users aren't using it. Number one, they're not competing on benchmarks. Number two, Humans don't prefer it.

Jordan Wilson [00:48:15]:
Number three. So can Claude recover? I don't know. I'd probably say no. Alright, y'all. I hope this was helpful. You wanted some hot takes? I tried to bring it. I tried to bring it a little bit. So, you know, talking a little bit has Anthropics Quad lost its edge.

Jordan Wilson [00:48:33]:
What happened and are Google and OpenAI too far ahead? Simple answer, yes. Anthropics lost its edge and, yes, OpenAI and Google, at least today, are way too far ahead for Anthropic to catch. I could be wrong, but the only way you're gonna find out is by continuing to tune in. Maybe I'll be eating a big helping of, you know, humble pie, you know, in 2026, but we will see and find out. Alright. Thank you for tuning in y'all. If you haven't already, please go to your everydayai.com. If this was helpful, please share this with your, network, tag a friend, someone that needs to hear this.

Jordan Wilson [00:49:10]:
If you're listening on the podcast, appreciate your support as always. Reach out to me. I always read my email, in my LinkedIn there, in the show notes. So please reach out if you have thoughts on this. You know, let me know in the livestream comments as well. Then go to your everydayai.com. Sign up for the free daily newsletter. Thanks for tuning in.

Jordan Wilson [00:49:28]:
We'll see you back tomorrow and every day for more everyday AI. Thanks, y'all.