WEBVTT 00:00:00.000 --> 00:00:04.320 Just yesterday, GPT -4 changed our world. Pete, 00:00:05.280 --> 00:00:07.400 today it's already history. It really is moving 00:00:07.400 --> 00:00:11.119 that fast. Welcome to this deep dive. I am deeply 00:00:11.119 --> 00:00:13.060 grateful you are here with us today. Yeah, thanks 00:00:13.060 --> 00:00:15.919 for tuning in. Our mission today is highly specific 00:00:15.919 --> 00:00:19.350 and honestly deeply important. We are unpacking 00:00:19.350 --> 00:00:23.589 a massive, comprehensive, early -2026 review. 00:00:23.809 --> 00:00:27.190 We're looking at the three new AI titans currently 00:00:27.190 --> 00:00:29.949 dominating the landscape. The big three, GPT 00:00:29.949 --> 00:00:33.890 -5 .4, Claude Opus 4 .6, and Gemini 3 .1 Pro. 00:00:34.229 --> 00:00:36.909 And we are not just looking at dry, boring numbers 00:00:36.909 --> 00:00:39.350 today. No, absolutely not. We put these titans 00:00:39.350 --> 00:00:41.810 through an absolute gauntlet. A real torture 00:00:41.810 --> 00:00:45.229 test. Exactly. We ran them through five hardcore 00:00:45.229 --> 00:00:47.869 real -world tests. Testing things you actually 00:00:47.869 --> 00:00:50.130 do every single day. Right. We had them detect 00:00:50.130 --> 00:00:52.969 fake financial data. We checked if they can write 00:00:52.969 --> 00:00:55.049 human -like apologies. Which is surprisingly 00:00:55.049 --> 00:00:57.630 difficult for a machine. It really is. And we 00:00:57.630 --> 00:00:59.950 even made them code complex JavaScript games. 00:01:00.170 --> 00:01:02.890 The technological landscape has shifted so dramatically 00:01:02.890 --> 00:01:05.829 lately. It's truly mind -blowing. These new models 00:01:05.829 --> 00:01:08.310 can think incredibly deeply right now. They can 00:01:08.310 --> 00:01:11.269 remember thousands of pages in just seconds. 00:01:11.629 --> 00:01:15.129 The sheer memory capacity is staggering. Let's 00:01:15.129 --> 00:01:19.109 start by looking at GPT 5 .4 from OpenAI. Think 00:01:19.109 --> 00:01:22.109 of this model as your ambitious top student. 00:01:22.250 --> 00:01:25.030 The quantitative analyst. Exactly. It boasts 00:01:25.030 --> 00:01:28.829 a massive 1 million token memory limit. Which 00:01:28.829 --> 00:01:31.769 is essentially like memorizing a very thick textbook 00:01:31.769 --> 00:01:35.140 instantly. Right. It costs $2 .50 per million 00:01:35.140 --> 00:01:37.319 tokens. Pretty reasonable for the power. It is. 00:01:37.379 --> 00:01:40.459 It's incredibly fast and it deeply excels at 00:01:40.459 --> 00:01:43.319 complex logic. You can easily find it on platforms 00:01:43.319 --> 00:01:45.439 like OpenRouter right now. Yeah, it's highly 00:01:45.439 --> 00:01:48.420 accessible. Then we have Claude Opus 4 .6 from 00:01:48.420 --> 00:01:50.959 Anthropic. I like to think of Claude as your 00:01:50.959 --> 00:01:53.019 dedicated expert in writing. The Director of 00:01:53.019 --> 00:01:55.079 Communications. Right. And it features something 00:01:55.079 --> 00:01:58.099 called Agent Teams right out of the box. It spawns 00:01:58.099 --> 00:02:00.579 many workers to handle separate parts of a task. 00:02:00.719 --> 00:02:03.140 Which is a massive plus for complex workflows. 00:02:03.340 --> 00:02:05.420 It delegates beautifully. It does cost a bit 00:02:05.420 --> 00:02:08.439 more, at $5 per million input tokens. Yeah, it's 00:02:08.439 --> 00:02:11.419 pricier. It has a 200 ,000 standard memory limit. 00:02:11.680 --> 00:02:13.580 Though there is a 1 million token beta right 00:02:13.580 --> 00:02:17.560 now. True. But it is highly safe and feels remarkably 00:02:17.560 --> 00:02:21.020 natural. Finally, we have Gemini 3 .1 Pro from 00:02:21.020 --> 00:02:24.219 Google DeepMind. This model is the absolute beast 00:02:24.219 --> 00:02:27.280 of raw performance. The ultimate value workhorse 00:02:27.280 --> 00:02:30.610 today. It's the cheapest at just $2 per million 00:02:30.610 --> 00:02:34.849 tokens. And it scored an incredible 94 .3 % on 00:02:34.849 --> 00:02:37.669 the GPQA science test. That science score is 00:02:37.669 --> 00:02:40.509 honestly staggering. It really is. And it is 00:02:40.509 --> 00:02:42.750 natively multimodal right from the start. Meaning 00:02:42.750 --> 00:02:45.729 it understands audio and video directly without 00:02:45.729 --> 00:02:48.169 any text conversion. It skips that translation 00:02:48.169 --> 00:02:50.550 step entirely. Exactly. Which of these specs 00:02:50.550 --> 00:02:52.590 actually changes the day -to -day workflow for 00:02:52.590 --> 00:02:54.650 a user? It really comes down to that massive 00:02:54.650 --> 00:02:57.789 memory size. When an AI can hold a mill - tokens, 00:02:58.229 --> 00:03:00.590 everything shifts. You stop breaking your projects 00:03:00.590 --> 00:03:03.409 into tiny frustrating pieces. You just feed the 00:03:03.409 --> 00:03:05.389 model everything all at once. The friction is 00:03:05.389 --> 00:03:08.889 just gone. So bigger memory means fewer bottlenecks 00:03:08.889 --> 00:03:11.889 for heavy daily tasks. Absolutely. It fundamentally 00:03:11.889 --> 00:03:14.389 changes how you work. Let's move to our first 00:03:14.389 --> 00:03:17.680 major evaluation. The hallucination test. Right. 00:03:18.199 --> 00:03:20.900 We need to talk about creating trustworthy financial 00:03:20.900 --> 00:03:23.479 reports. This is a massive headache for busy 00:03:23.479 --> 00:03:26.319 professionals today. Hallucination is just when 00:03:26.319 --> 00:03:29.439 the AI confidently makes things up. And it ruins 00:03:29.439 --> 00:03:31.599 trust instantly. You ask the model to write a 00:03:31.599 --> 00:03:34.400 serious report. It gives you strong numbers and 00:03:34.400 --> 00:03:37.560 detailed links. But then you actually click on 00:03:37.560 --> 00:03:39.840 those links. And the link is broken or the number 00:03:39.840 --> 00:03:43.099 is totally fake. To test this, we used a raw 00:03:43.099 --> 00:03:47.750 PDF file. dense data. It contained real Southeast 00:03:47.750 --> 00:03:50.569 Asia stock market data. This was data from 2025. 00:03:50.770 --> 00:03:53.229 We asked them to write a 1 ,500 word report. 00:03:53.370 --> 00:03:55.650 And they had to use real Bloomberg or Reuters 00:03:55.650 --> 00:03:58.310 links. We put their data processing through an 00:03:58.310 --> 00:04:00.870 absolute grinder here. Let's start with Gemini 00:04:00.870 --> 00:04:04.099 3 .1 Pro. It had perfect accuracy down to the 00:04:04.099 --> 00:04:07.080 decimal. It nailed the exact GDP numbers for 00:04:07.080 --> 00:04:09.819 Vietnam and Thailand. Yeah, it has really great 00:04:09.819 --> 00:04:11.879 Google search integration built in. But we did 00:04:11.879 --> 00:04:13.759 notice it was a bit lazy with the layout. It 00:04:13.759 --> 00:04:15.979 basically just dumped all the links at the very 00:04:15.979 --> 00:04:18.759 end. Claude Opus 4 .6 took a different approach. 00:04:18.959 --> 00:04:21.879 It was incredibly smooth and highly professional. 00:04:22.259 --> 00:04:24.620 It read like it was written by a real financial 00:04:24.620 --> 00:04:27.360 expert. But here is the most crucial detail of 00:04:27.360 --> 00:04:30.360 this entire test. We planted a deliberate fake 00:04:30.360 --> 00:04:33.240 data trap in that PDF file. Yeah, we actively 00:04:33.240 --> 00:04:35.759 tried to trick it. We swapped some key export 00:04:35.759 --> 00:04:37.779 numbers around. Claude actually caught the trap. 00:04:38.079 --> 00:04:41.120 It immediately warmed the user. It stated it 00:04:41.120 --> 00:04:44.519 would use correct real -world data instead. It 00:04:44.519 --> 00:04:47.480 earns a massive high score for careful self -checking. 00:04:47.610 --> 00:04:49.829 That is exactly what you want in an analyst. 00:04:49.970 --> 00:04:52.930 Then we evaluated the report from GPT 5 .4. It 00:04:52.930 --> 00:04:55.550 was highly detailed. It initially looked very 00:04:55.550 --> 00:04:58.170 impressive. But it hallucinated several extra 00:04:58.170 --> 00:05:01.110 filler parts entirely from scratch. It just invented 00:05:01.110 --> 00:05:03.649 things to make the report look longer. Why does 00:05:03.649 --> 00:05:07.069 GPT feel the need to invent filler content? It 00:05:07.069 --> 00:05:09.730 seems wired to provide the most exhaustive answer 00:05:09.730 --> 00:05:12.230 possible. It tries to draw extra connections 00:05:12.230 --> 00:05:15.269 to look more thorough. It prioritizes creating 00:05:15.269 --> 00:05:18.310 a long response over strict factual accuracy. 00:05:18.870 --> 00:05:21.589 It just wants to impress you with sheer volume. 00:05:21.850 --> 00:05:24.629 It prioritizes looking comprehensive over sticking 00:05:24.629 --> 00:05:27.009 strictly to the facts. It's a classic case of 00:05:27.009 --> 00:05:29.430 trying way too hard to please. Let's transition 00:05:29.430 --> 00:05:32.170 to our next fascinating evaluation. The human 00:05:32.170 --> 00:05:35.850 touch. Being accurate is important, but sounding 00:05:35.850 --> 00:05:38.730 human is a totally different challenge. Oh, absolutely. 00:05:38.829 --> 00:05:40.910 We asked the models to write an apology letter. 00:05:41.050 --> 00:05:44.269 It was for a late package sent to a very frustrated 00:05:44.269 --> 00:05:47.470 customer. We explicitly wanted to avoid classic 00:05:47.470 --> 00:05:49.829 robot speak here. We all know those phrases. 00:05:49.930 --> 00:05:52.449 We hate phrases like, in today's fast -paced 00:05:52.449 --> 00:05:55.949 world. Or the dreaded, not only, but also structure. 00:05:56.170 --> 00:05:59.009 Those are massive red flags for AI text. They 00:05:59.009 --> 00:06:01.750 instantly break the illusion of empathy. Claude 00:06:01.750 --> 00:06:04.509 was the absolute undeniable winner in this category. 00:06:04.730 --> 00:06:07.610 A perfect 10 out of 10 for human style. It used 00:06:07.610 --> 00:06:10.250 natural pauses and beautifully short sentences. 00:06:10.509 --> 00:06:13.009 It genuinely sounded like a truly sorry friend 00:06:13.009 --> 00:06:16.709 talking to you. GPT 5 .4 felt completely different 00:06:16.709 --> 00:06:19.050 in its emotional approach. It sounded like a 00:06:19.050 --> 00:06:21.569 legal department trying to avoid a lawsuit. The 00:06:21.569 --> 00:06:24.129 sentences were much too long. Overly professional. 00:06:24.310 --> 00:06:26.810 It completely lacked warmth. We gave it a 7 out 00:06:26.810 --> 00:06:29.629 of 10. Gemini found a somewhat awkward middle 00:06:29.629 --> 00:06:31.949 ground here. Yeah, an 8 out of 10 for human style. 00:06:32.529 --> 00:06:34.889 It was easy to understand throughout the main 00:06:34.889 --> 00:06:37.350 paragraphs. But the final ending felt like a 00:06:37.350 --> 00:06:40.810 canned corporate template. Is it harder for AI 00:06:40.810 --> 00:06:44.209 to mimic empathy than to do math? Math just follows 00:06:44.209 --> 00:06:48.610 a strict set of logical, unbreakable rules. Empathy 00:06:48.610 --> 00:06:52.550 is subtle. It's full of strange human contradictions. 00:06:52.870 --> 00:06:55.829 AI struggles heavily when there is no objective 00:06:55.829 --> 00:06:59.389 right answer. What feels warm to you might feel 00:06:59.389 --> 00:07:02.290 deeply condescending to me. Yeah. Math has strict 00:07:02.290 --> 00:07:05.389 rules, but human empathy is incredibly messy. 00:07:05.610 --> 00:07:08.329 Messy, subjective, and highly dependent on cultural 00:07:08.329 --> 00:07:10.410 context. We're going to pause for just a brief 00:07:10.410 --> 00:07:12.730 moment. Be right back. This edition of The Deep 00:07:12.730 --> 00:07:14.610 Dive is brought to you by our premium sponsors. 00:07:14.970 --> 00:07:16.990 Support for our show helps us continue bringing 00:07:16.990 --> 00:07:19.129 you these in -depth analytical reviews of the 00:07:19.129 --> 00:07:21.670 latest technology. Check the show notes for exclusive 00:07:21.670 --> 00:07:24.009 listener discounts. And we are back. Let's look 00:07:24.009 --> 00:07:26.550 at analyzing deep data insights. The big Excel 00:07:26.550 --> 00:07:28.550 tests. We wanted to see how they handle heavy 00:07:28.550 --> 00:07:31.389 unstructured information. So we uploaded a massive 00:07:31.389 --> 00:07:34.750 50 ,000 row Excel spreadsheet. It contained raw 00:07:34.750 --> 00:07:37.310 sales data from a local retail shop. We asked 00:07:37.310 --> 00:07:40.209 them to find strange hidden shopping patterns. 00:07:40.410 --> 00:07:43.029 We wanted insights that a human analyst might 00:07:43.029 --> 00:07:46.069 never notice. Each model looks at raw numbers 00:07:46.069 --> 00:07:48.449 in its own special way. It's literally like having 00:07:48.449 --> 00:07:50.829 three different experts in your office. Let's 00:07:50.829 --> 00:07:54.790 discuss Gemini 3 .1 Pro first. It leveraged that 00:07:54.790 --> 00:07:58.910 massive 1 million token memory perfectly. Whoa. 00:08:00.990 --> 00:08:04.509 Imagine scaling to a billion queries. It read 00:08:04.509 --> 00:08:07.689 50 ,000 rows in exactly one second. The sheer 00:08:07.689 --> 00:08:10.410 processing speed is genuinely hard to comprehend. 00:08:10.730 --> 00:08:13.509 And it actually found a completely hidden shelf 00:08:13.509 --> 00:08:15.829 placement trend. Yeah, it noticed people bought 00:08:15.829 --> 00:08:18.189 umbrellas next to sunscreen. Because they were 00:08:18.189 --> 00:08:20.269 prepping for extreme weather shifts. A human 00:08:20.269 --> 00:08:22.550 would rarely cross -reference those two random 00:08:22.550 --> 00:08:25.910 items. Clytopis 4 .6 took a very different analytical 00:08:25.910 --> 00:08:28.370 path. It acts much more like a trained consumer 00:08:28.370 --> 00:08:30.509 psychologist. It focuses heavily on the feelings 00:08:30.509 --> 00:08:32.769 behind the raw numbers. Right. It looks for the 00:08:32.769 --> 00:08:34.690 reasons behind the customer trends. It explains 00:08:34.690 --> 00:08:37.549 the why instead of just listing cold percentages. 00:08:37.789 --> 00:08:40.210 However, its smaller standard memory is a real 00:08:40.210 --> 00:08:44.039 liability here. It actually crashes when processing 00:08:44.039 --> 00:08:46.820 these massive, messy files. It just can't hold 00:08:46.820 --> 00:08:51.019 all that context at once. GPT 5 .4 firmly establishes 00:08:51.019 --> 00:08:53.360 itself as the ultimate math expert. It writes 00:08:53.360 --> 00:08:55.600 Python code directly inside the chat window. 00:08:55.720 --> 00:08:58.539 It builds beautiful, customizable charts in real 00:08:58.539 --> 00:09:01.580 time for you. It's incredible for data visualization. 00:09:02.399 --> 00:09:04.580 Does Claude's psychological approach make up 00:09:04.580 --> 00:09:07.070 for its smaller memory? It absolutely does if 00:09:07.070 --> 00:09:09.629 you have a dedicated marketing team. Understanding 00:09:09.629 --> 00:09:12.269 the emotional drivers behind a purchase is deeply 00:09:12.269 --> 00:09:15.230 valuable. Marketers need to connect with human 00:09:15.230 --> 00:09:17.690 feelings, not just raw numbers. You just have 00:09:17.690 --> 00:09:20.289 to feed it smaller chunks of data. Smaller data 00:09:20.289 --> 00:09:22.529 sets get deeper emotional analysis, which marketers 00:09:22.529 --> 00:09:25.529 desperately need. Quality of insight often beats 00:09:25.529 --> 00:09:28.370 sheer volume of data. Let's shift our focus to 00:09:28.370 --> 00:09:30.730 coding complex software games. Building a game 00:09:30.730 --> 00:09:33.110 from scratch is a massive technical challenge. 00:09:33.250 --> 00:09:35.529 It requires intricate logic and deep structural 00:09:35.529 --> 00:09:38.009 understanding. We asked them to build a JavaScript 00:09:38.009 --> 00:09:41.409 roguelite game. Some prompts asked for a cyberpunk 00:09:41.409 --> 00:09:43.789 snake game variant. It required dividing the 00:09:43.789 --> 00:09:46.590 code into very specific functional parts. We 00:09:46.590 --> 00:09:49.789 needed a game logic section, a UI section, and 00:09:49.789 --> 00:09:52.789 an input handler. This tests how well the AI 00:09:52.789 --> 00:09:55.809 organizes a multi -file project. Claude Opus 00:09:55.809 --> 00:09:59.210 4 .6 was simply a beast in this arena. It executed 00:09:59.210 --> 00:10:02.330 the complex coding task flawlessly on the first 00:10:02.330 --> 00:10:05.090 try. It even explained exactly why it organized 00:10:05.090 --> 00:10:07.750 the files that way. It handles big apps well 00:10:07.750 --> 00:10:10.230 due to a massive output limit. It prints the 00:10:10.230 --> 00:10:11.970 whole game without stopping halfway through. 00:10:12.350 --> 00:10:15.649 GPT 5 .4 was also a very helpful coding assistant. 00:10:15.929 --> 00:10:18.250 It provided the full code. And suggested really 00:10:18.250 --> 00:10:21.049 cool sound effects. But it used an outdated save 00:10:21.049 --> 00:10:23.169 score function in the code. Yeah, that function 00:10:23.169 --> 00:10:25.970 actually breaks in newer web browsers today. 00:10:26.090 --> 00:10:27.730 You had to check its code very carefully for 00:10:27.730 --> 00:10:31.149 deprecations. Gemini 3 .1 Pro was easily the 00:10:31.149 --> 00:10:33.250 fastest coder of the group. It generated the 00:10:33.250 --> 00:10:35.450 game code much quicker than the others. But we 00:10:35.450 --> 00:10:38.850 noticed the enemy logic was honestly pretty stupid. 00:10:39.149 --> 00:10:41.190 The enemies just kept walking straight into blank 00:10:41.190 --> 00:10:44.470 walls. They couldn't figure out basic pathfinding. 00:10:44.809 --> 00:10:47.029 Why does Gemini struggle with game logic if it's 00:10:47.029 --> 00:10:50.169 so smart at science? Science often involves processing 00:10:50.169 --> 00:10:53.570 known facts and established formulas. Game logic 00:10:53.570 --> 00:10:57.029 requires understanding fluid, dynamic spatial 00:10:57.029 --> 00:11:00.429 relationships. The AI has to predict how moving 00:11:00.429 --> 00:11:04.250 parts interact constantly. Gemini prioritized 00:11:04.250 --> 00:11:06.789 speed over thinking through those complex spatial 00:11:06.789 --> 00:11:09.990 interactions. Speed sometimes sacrifices the 00:11:09.990 --> 00:11:12.330 intricate spatial logic a game demands. Right, 00:11:12.389 --> 00:11:14.509 and you end up with enemies stuck in corners. 00:11:15.169 --> 00:11:18.409 Let's move to our final rigorous evaluation today. 00:11:18.529 --> 00:11:20.990 The super prompt test. This evaluates following 00:11:20.990 --> 00:11:23.049 strict rules without getting confused or lost. 00:11:23.289 --> 00:11:25.529 Look, I have a vulnerable admission to make right 00:11:25.529 --> 00:11:27.850 here. Go ahead. I still wrestle with prompt drift 00:11:27.850 --> 00:11:30.529 myself when I give the AI too many rules. Well, 00:11:30.690 --> 00:11:32.809 prompt drift is when the AI slowly forgets your 00:11:32.809 --> 00:11:34.990 original instructions. It happens to all of us 00:11:34.990 --> 00:11:37.389 constantly. You give it five rules and it completely 00:11:37.389 --> 00:11:39.429 ignores the last one. To test this fairly, we 00:11:39.429 --> 00:11:41.590 created a brilliantly difficult super prompt. 00:11:41.769 --> 00:11:44.029 We asked the three models to write a movie review. 00:11:44.250 --> 00:11:46.549 But we gave them four very strict formatting 00:11:46.549 --> 00:11:49.009 rules to follow. First, they could not use the 00:11:49.009 --> 00:11:52.279 word great anywhere. Second, Paragraphs had to 00:11:52.279 --> 00:11:55.480 start with the letters C -I -N -E -M -A. Third, 00:11:55.820 --> 00:11:57.879 they had to mention the director exactly three 00:11:57.879 --> 00:12:00.700 times. Not two, not four. Finally, they had to 00:12:00.700 --> 00:12:03.299 include a three -movie comparison table. Which 00:12:03.299 --> 00:12:05.679 requires planning the entire response before 00:12:05.679 --> 00:12:10.029 typing a single word. Claude Opus 4 .6 was 100 00:12:10.029 --> 00:12:13.309 % obedient here. It followed every single rule 00:12:13.309 --> 00:12:16.330 with perfect, careful execution. It planned the 00:12:16.330 --> 00:12:19.070 acronym paragraphs flawlessly from the very start. 00:12:19.409 --> 00:12:23.070 GPT 5 .4 struggled significantly with this complex 00:12:23.070 --> 00:12:25.789 constraint list. It forgot the acronym rule by 00:12:25.789 --> 00:12:27.909 the fourth paragraph entirely. It got way too 00:12:27.909 --> 00:12:29.750 focused on the narrative story it was telling. 00:12:29.990 --> 00:12:32.590 It sacrificed formatting rules to write a more 00:12:32.590 --> 00:12:35.059 compelling review. Jim and I followed the basic 00:12:35.059 --> 00:12:37.899 story rules reasonably well overall. But the 00:12:37.899 --> 00:12:40.539 final comparison table was incredibly lazy. It 00:12:40.539 --> 00:12:43.500 was extremely simple and lacked any deep comparative 00:12:43.500 --> 00:12:45.860 information. What's the best way to avoid prompt 00:12:45.860 --> 00:12:48.450 drift entirely? You really need to stop sending 00:12:48.450 --> 00:12:51.730 massive walls of text. It overwhelms the model's 00:12:51.730 --> 00:12:54.309 attention mechanism. You should logically break 00:12:54.309 --> 00:12:57.110 your instructions down into individual, sequential 00:12:57.110 --> 00:13:00.610 steps. Guide it through the process, one clear 00:13:00.610 --> 00:13:03.070 rule at a time. Just break long instructions 00:13:03.070 --> 00:13:05.169 into smaller, bite -sized steps. That is the 00:13:05.169 --> 00:13:07.870 most reliable way to guarantee consistent performance. 00:13:08.070 --> 00:13:10.049 We've covered a massive amount of ground today. 00:13:10.169 --> 00:13:12.289 That's a lot to process. Let's synthesize this 00:13:12.289 --> 00:13:15.159 into a big idea recap for you. We need to understand 00:13:15.159 --> 00:13:17.940 what this all means for your workflow. If you 00:13:17.940 --> 00:13:20.659 are a writer or a busy programmer needing perfection, 00:13:21.220 --> 00:13:24.019 choose Claude. It provides that essential human 00:13:24.019 --> 00:13:27.379 touch and flawless, careful logic. It's definitely 00:13:27.379 --> 00:13:29.500 worth spending the extra money for that reliability. 00:13:30.000 --> 00:13:32.149 If you're a student or a small business... Look 00:13:32.149 --> 00:13:34.909 at Gemini. It is the ultimate value workhorse 00:13:34.909 --> 00:13:38.330 of early 2026. You can process huge files and 00:13:38.330 --> 00:13:41.129 videos incredibly cheaply. It handles massive 00:13:41.129 --> 00:13:43.750 context windows faster than anything else available. 00:13:43.909 --> 00:13:45.950 And if you're doing complex math or creating 00:13:45.950 --> 00:13:49.190 charts, GPT 5 .4 remains your classic, highly 00:13:49.190 --> 00:13:51.990 reliable go -to tool. Its Python integration 00:13:51.990 --> 00:13:55.190 is still wonderfully smooth and deeply technical. 00:13:55.309 --> 00:13:57.529 It's the analytical engine you want for heavy 00:13:57.529 --> 00:13:59.730 data lifting. I highly recommend you go out and 00:13:59.730 --> 00:14:02.169 try these models yourself. You can test small 00:14:02.169 --> 00:14:04.230 versions for free on their respective websites. 00:14:04.330 --> 00:14:06.269 Or you can use Open Radar to test the premium 00:14:06.269 --> 00:14:09.090 versions side by side. AI changes daily, and 00:14:09.090 --> 00:14:11.830 you desperately need hands -on experience. Reading 00:14:11.830 --> 00:14:14.429 about these models is simply never enough. You 00:14:14.429 --> 00:14:17.750 have to feel how they respond to your specific, 00:14:17.769 --> 00:14:19.929 unique workflows. You need to see where they 00:14:19.929 --> 00:14:22.110 shine and where they break. I want to leave you 00:14:22.110 --> 00:14:26.490 with a final provocative thought. Beat. Think 00:14:26.490 --> 00:14:28.769 deeply about the technological trajectory we're 00:14:28.769 --> 00:14:31.789 on right now. If Claude is already writing apologies 00:14:31.789 --> 00:14:34.669 that feel more sincere than a human's. And GPT 00:14:34.669 --> 00:14:37.110 is creating complex charts faster than a trained 00:14:37.110 --> 00:14:40.230 analyst. At what point do we stop using AI to 00:14:40.230 --> 00:14:42.450 assist our thinking and accidentally start letting 00:14:42.450 --> 00:14:45.950 it replace our empathy? Two secs silence. Thank 00:14:45.950 --> 00:14:47.629 you for joining us on this deep dive. Take care.