процессов или обработки конфиденциальных данных, имеющих большое значение. Кроме того, они внедряют кэширование с мгновенным откликом для повышения эффективности, а их SDK позволяют создавать собственные агенты для кодирования. Теперь Claude Code напрямую интегрируется с популярными инструментами разработки, такими как VS Code и Jet Brains. Иными словами, они предпринимают скучные, но крайне важные шаги, чтобы убедиться, что эта технология не только
Поиск по транскрипциям
this demo, I didn't run into that at all. So, that's very nice. But I'd say the biggest difference between Gemini 3 at the moment and Claude is Gemini just has a lot of rough edges where you ask it to do some things and it doesn't do it perfectly. So it's going...been using anti-gravity? Let me know down in the replies below. I also just launched the Vibe Coding Academy. It's my private community where has a full Claude code master class as well as weekly meetings with myself where I answer your questions. I help guide you through building apps. Link for that is down below. Check that
Haiku stacks up with software engineering. It is right behind Sonnet 45, which I think is the best coding model ever made. I have gotten such unbelievable results in claude code using Sonnet 45. So, it is right behind Sonnet 45, which is actually mind-blowing because it is a third of the price of Sonnet 45. It is stronger...release a new model soon. We've been on 2. 5 Pro for like 4 years now in AI term. Looking at some other benchmarks here from Claude Haiku, it looks like when it comes to a gentic tool use. So just as a straightup AI agent doing things on your computer, it is ahead of the competition but still
essentially which AI can not just get the search, which is useful, and it seems like Claude does the most comprehensive searches cuz it searches three different times, but also the difference here is that if you ask it to then code off the back of that, look at this. Right? So, if we say create a landing page website...another breakdown, right? That is not a coded website. It doesn't give us a landing page. It's pretty basic and it doesn't code anything. If we go back to Claude now, Claude 3. 7, it will actually go off and it's intelligent enough to go, "Ah, this person asked for a website, so I'm going
Nice effects. I like the background of all the stars and everything like that. It's looking pretty cool. Gemini looks just about finished. Deep Seek still coding. And we got Claude ready to go as well here. Claude, you can actually control the game with your mouse. it looks like. And here's a shoot button. I would...versus Claude 4 versus Gemini 2. 5 Pro. I just ran China's new version of Deep Sea Garwan against Claude 4 and Gemini 2. 5 Pro and the results have shocked me. This is a brutal AI coding battle and what happened next is going to be very interesting. I'm about to show you which AI destroys
logic and reasoning of these models actually ends up touching. - Ultimately, these are digital systems that we interact with every single day, right? The fact that Claude is great at code gives it a flexible skill and a shortcut to do all of these really cool, interesting things, right? Our file creation feature that was launched a few weeks...that enables Claude to create Excel documents and PowerPoint is essentially Claude accessing a virtual machine within which it can run Python code at scale
coding task creating a landing page based on search results Claude understood exactly what I wanted and created a beautiful...functional website with clear navigation and well structured content chat GPT completely missed the point and just gave me some text about websites instead of actually coding one it's clear that Claude with web search is now more powerful than ever it was already the best model for writing and coding and now with access to upto-date information
through this multiplication and uh do it in its head and get it wrong but actually like remarkably close uh then I tried Claud and Claude actually wrote In this case not python code but it wrote JavaScript code but uh JavaScript is also a programming l language and get gets the correct result then I came to Gemini
well at all, which is really weird. But yeah, that's coming in last so far. And let's just test what we got back from Claude here. Just give me the code, mate. I don't want to run it. Where's Why is it Give me the code. There we go. All right, let's grab that...work properly. You can see the background, etc. It's not that nice, but it comes in second cuz it does actually work unlike Claude's output. So, just to recap here on the coding games test, I'm saying that GPT4. 1 is winning by quite a long way here. and side by side it's winning on everything
Just type terminal and copy this line here. And all you got to do is paste a line here and it should install. And then just type claude to load claw code. Okay. So there you go. Clock code is now live in terminal. But personally, you know, I prefer to use claw code inside an IDE like cursor...days instead of using the cursor agent, I prefer to just open terminal incursor here and type claude and say, you know, yes, I trust these files. And that's pretty much all it takes. And now you can use claw code in cursor. And the reason I like using clock code in cursor is because you can easily view
monster just dropped and it's already making waves. This thing is their latest coding powerhouse that's designed to compete with the best. Then we have Claude for Opus, Anthropics coding beast. They claim it's the world's best coding model. It can code for 7 hours straight without stopping. And that's like having a developer...start screen, score tracking, increasing difficulty over time, and work well on both desktop and mobile. I asked each AI to build a complete game, not just some basic code, a real game with all these features. This is where things got interesting. Gemini came out swinging, super fast response. The game actually worked. You could play it right away
monster just dropped and it's already making waves. This thing is their latest coding powerhouse that's designed to compete with the best. Then we have Claude for Opus Anthropic coding beast. They claim it's the world's best coding model. It can code for 7 hours straight without stopping. And that's like having a developer...start screen, score tracking, increasing difficulty over time, and work well on both desktop and mobile. I asked each AI to build a complete game, not just some basic code, a real game with all these features. This is where things got interesting. Gemini came out swinging, super fast response. The game actually worked. You could play it right away
monster. Just dropped and it's already making waves. This thing is their latest coding powerhouse that's designed to compete with the best. Then we have Claude for Opus Anthropics coding beast. They claim it's the world's best coding model. It can code for 7 hours straight without stopping. That's like having a developer who never...start screen, score tracking, increasing difficulty over time, and work well on both desktop and mobile. I asked each AI to build a complete game, not just some basic code, a real game with all these features. This is where things got interesting. Gemini came out swinging, super fast response. The game actually worked. You could play it right away
monster just dropped and it's already making waves. This thing is their latest coding powerhouse that's designed to compete with the best. Then we have Claude for Opus, Anthropics coding beast. They claim it's the world's best coding model. It can code for 7 hours straight without stopping. And that's like having a developer
about real world performance. But here's what's really crazy. When I dug into the actual benchmark data, the story gets way more complicated. Yes, Claude Opus 4 dominates the swbench coding test with 72. 5%. But check this out. On live codebench, Gemini Deepth think actually scores 87. 6%. That destroys Claude's performance. And in math, Gemini...medal grade. But here's the kicker. Claude still wins on Terminal Bench with 43. 2%. Gemini, no delta available. Here's another shocker. Claude can work for up to seven hours straight on complex coding tasks without losing focus. Seven hours. That's like having a dedicated programmer who never gets tired, never makes mistakes, and never asks
yourself every single session. This changes everything for anyone building with AI. Let me show you why this is insane. All right, so there's a new integration for Claude Code that's genuinely mind-blowing. It's called Super Memory, and it fixes the biggest problem everyone has with AI coding tools. Let me explain. You know that thing...starting from zero. Claude forgot everything. You have to explain it all over again. It's like Groundhog Day, but for coding. I used to keep these massive files just to paste into Claude every time just so it would remember my setup. It was exhausting. And I know I'm not the only one. Every developer I talk
just to recap here, if I had to pick one so far, Gemini, I would say, is probably performing the best. Deepseek is creating good stuff as well. Claude as well is up there, but Chat GPT is not even in the race at this point. So, Chat GPT03 finished first here. Let's test this out. Here...Nice effects. I like the background of all the stars and everything like that. It's looking pretty cool. Gemini looks just about finished. Deep Seek still coding. And we got Claude ready to go as well here. Claude, you can actually control the game with your mouse, it looks like. And here's a shoot button. I would
than 50 LM. Yeah, I heard Amazon Bedrock is pretty good. " Actually, the new Claude update for the million tokens context window that was only available in two places. Claude Code and Amazon Bedrock. It wasn't available anywhere else. That dude says, "What LLM will allow us to schedule an agent to dedicate a task? " I can't remember...currently the best for coding? I really love Gemini. If you're building something basic, even more complex, then I would say Claude or Gemini, but Gemini tends to have it's got three APIs for coding, so you know, you can't argue with that. DT says I find Plexi the best scheduling. Dot says coding debugging. Oh yeah
than 50 LM. Yeah, I heard Amazon Bedrock is pretty good. " Actually, the new Claude update for the million tokens context window that was only available in two places. Claude Code and Amazon Bedrock. It wasn't available anywhere else. That dude says, "What LLM will allow us to schedule an agent to dedicate a task? " I can't remember...currently the best for coding? I really love Gemini. If you're building something basic, even more complex, then I would say Claude or Gemini, but Gemini tends to have it's got free APIs for coding, so you know, you can't argue with that. DT says I find Plexi the best scheduling. Dot says coding debugging. Oh yeah
then you have different AI models to choose from. Even though this is designed by Google, they actually give you access to other coding large language models too. So you have access to claude and GPTOSS which is this is the open source model from OpenAI here...upgrade with the other plans, you get even more access to things. As of right now, Gemini 3 Pro is one of the best AI coding models available and so is Claude Opus 4. 5. And you'll see inside of anti-gravity, you'll see different Gemini 3 Pro models. So you'll see one that says