NEW Claude 3.7 Sonnet Is Simply The Best Coding AI (Claude Code Testing)
Making apps has never been easier. Join the community for personalized help with getting started:
Claude 3.7 Sonnet: The BEST Coding LLM Ever! (Fullly Tested) – TRULY INSANE!
Links:
Prompt:
Build a personal finance tracker using Python and a web-based dashboard. Users should be able to log expenses, categorize spending, and visualize trends with Matplotlib. The app should include an AI feature that suggests ways to save money based on spending patterns.
Chapters:
0:00 Start
#claude #37sonnet #claudecoder
Free AI Resources:
🔑 Get My Free ChatGPT Templates:
🌟 Receive Tailored AI Prompts + Workflows:
👑 Explore Curated AI Tool Rankings:
🐦 Twitter:
📸 Instagram:
Premium Options:
🎓 Join the AI Advantage Courses + Community:
🛒 Discover Work Focused Presets in the Shop:
Forgot to mention 1 essential part: It’s a thinking model! That’s why it’s so damn good.
Also to clarify the lricing: Sonnet 3.7 = free to use in claude.ai (with limits), Claude Coder = free to use in terminal BUT you pay per prompt as you need to connect your API key
Alright now go and try it for yourself on the free plan and build something.
Actually, 03-mini-high was the first app that managed to help me setup a COT setup to generate reports, use external API to get stats and datas. And NOT use COT to analyze the report. All in 03-Mini-High. I do find that Claude 3.5 is much better than 03-Mini-High. I can’t imaging what Claude 3.7 can do. So I don’t really need the $200 package! Just the apps being created alone is a whole lot. Practical applications!!
what is an API key?
@@blueizclues2697 Protip: ask an AI.
You said it right. It’s all about the taste. With all the hype Claude gets, even though it is below other frontier models.
I get the point 😺
I have been using Claude code, and it’s incredible. In an agenetic workflow, I’ve created a WordPress plugin with a complete color system that looks like a Pantone style book in less than 15 minutes. It’s crazy.
Oh, but it’s eating through my credits like crazy. I mean, I spent $5 already just playing around with it in one session.
Hey Max, cool seeing you out in the wild. You’ll probably remember me from the Cwicly forums.
What’s your WP tools and workflow these days?
@@orrettwynter6731 Cursor, Pinegrow, Bricks…
Both 3.7 and 3.7T are in cursor already. Well done everyone
🤣 they wasted no time! I’m going to check rn
@EDashMan I was using 3.7T all day. Love it.
If you stop to think how fast AI is accelerating it’s actually insane. We are actually in an exponential guys.
In the midst of an intelligence explosion 💥
All the competition is good. It’s pushing these AI companies to keep improving their product at a faster rate.
@@sanseverything900 exactly! And nothing like an adversarial relationship between the US and China to push progress into overdrive; think the Russia v. US space race.
Earlier, I was keeping a regular check on the developments…now I think I should check back after 1 yr and we might be at a very very different landscape and the tasks will be done in less number of steps and much quicker than now
Yeah, this stuff is scary and exciting for me. I don’t know what to do as a developer now except to go all-in on AI code tooling and hope I maintain some sort of marketable edge due to actually being a developer (whatever that might entail, whether it’s being able to adequately review AI code, prompt in a more knowledgeable and streamlined manner, or something else altogether).
I used Claude 3.7 on Cursor to make a rudimentary version of DOOM in Python.
It took me less than an hour to do and I didn’t even have to write a single line of code except for the one time I had to manually edit the mouse sensitivity.
Otherwise, the game was flawless in terms of what I was trying to achieve.
try to make a simple version of Minecraft
Doom game? Did it also generate sprites for you? Otherwise I don’t get how it would be Doom?
@@D03t0nI’ll try that next when I get the time!
@@UnchartedWorlds I told it to make enemies and to allow me to shoot them and have them attack me when they spotted me. The enemies were green cylinders lol
Again, it’s rudimentary at best.
@@Arepo2801 that’s gonna take a while, and alot of errors and fixing
Used Claude Code today on a hobby project and it was awesome. Gonna be my go-to tool at least whenever I’m fine with the extra API costs compared to DeepSeek.
I was still using claude-3.5 in cursor because it was the only model that worked the best for me. 4o made a lot of mistakes and deepseek r1 was, ironically, overthinking problems and it made mistakes too. o3 was taking too long. Of course 3.5 wasn’t perfect either but it was the best performance and accuracy wise at least for me. Super excited about 3.7!
Thanks Igor for letting us know of yet another new release. Hard to keep up.
LOL after step 2,485. Your so right, I have always loved Claude and my experiment with Grok is over. Anthropic treats it’s people amazingly well, they are more concerned about the user experience and not winning in the dumb benchmarks. This company is the real deal!
The best thing about Claude Code is that others will be falling over themselves to create a similar tool as soon as possible.
In the future, AI will build personalized apps and websites for you, taking care of security and other technical aspects. This means everyone can have their own custom system without needing to code
this year 😉 soon come
In the future, AI will be able to assist you in coding, but if youre dum as fk like you then you’ll write mediocre code already available on the web from it. Otherwise if your brain is working then AI will just help you be faster
We already have that, its just that people don’t know the how/why yet. If you showed someone an iphone in 1982 they would have no idea. Its here.
As a software engineer, I think Claude has always been the best all around model. BUT ai only gets so far before it fails completely and cost more time than writing code on your own. In production software, the project is huge and across different architectures and languages. It would be impossible to feed all those source code into ai and try and get some work done. I have been using ai for coding since the chatgpt 3 beta public release. It is improving but never perfect.
The problem with software engineering is that a source code needs to be “perfect” in terms of correctness before it can even run. Since ai always will have a limited context length and can not “unlearn” or summarize the previous contexts, ai most likely will only be used as a tool and not replace an engineer. The generations are always limited and one time only. All you needed is a website with python backend then it will get it done easily. But once you need to change the code or improve it, it becomes impossible unless you understand what is happening and write your own code.
A recurring theme I’d been noticing is that every time I say “..yeah but it could never…” about AI I realize that if a human or a group of humans can collectively figure it out then, in time, without fail – so will ai.
@ Give one example that ai replaced human. There isn’t even one example that ai can replace humans. Ai doesn’t “figure out” anything, it is an inefficient compression algorithm that takes the greatest common divisor of things you put in it. If you tried to use ai art, you know it is all great looking, but can’t be used in real projects without looking same or without real human edits. In software, ai is a great way to learn topics that are widely known, but fails to do real work for you (in a large code base adding features, refactoring code, and changing code has being really hard for ai to do). If all you need is an isolated small, and average software, then ai will do it perfectly.
@the-programing you completely miss what’s being said in these model releases. They are now at 0.8 % effective. Not even 1 percent. Back in February 26th 2024, it was 0.04%. Anthropic deepseek and other llm producers say they are IMPROVING. improving. Stop being in denial. Claude 3.7 is now at 2.3% as good as senior software engineer. Buy September 2025, llm!s will be at 8 to 14 %as good. Prepare early or get swept out
@@ShmuelLevinson As a software engineer, that’s not how it works at all.
IE, this video uses the correct terminology, code generation. Code generations is exactly that, only code generation, it does not test the code, it is usually not aware of any recent updates, bug fixes, it can’t even tell if the code it produced it even correct.
Can it generate code? Yes.
But it’s a far cry away from replacing engineers. Actually they recently released an experiment where they replaced an entire junior dev team for AI, and it was a HUGE failure. AI in the right hands can be guided to right solution, however, if the developer has no idea what they are doing, it still produces garbage code.
If you teach a monkey to build a motorcycle it can build a motorcycle, but only that brand and type, it still can’t tell you what the problem is if something comes up.
Overall it’s great at producing what is called scripts, not working on whole projects. It’s tool and not yet a replacement any developer.
As a former military coder, in assembly I always end the day saying: This is so 2024 🙂
You have not checked how many tokens one request uses. And it uses a lot of them. According to estimates, 1 hour of work in Claude Code on the average project will cost $ 5-10 at least. Isn’t it much? Guys, open your wallets wider! 😄😄😄
Everyday “the best” new something comes out
welcome to ai in 2025
In the near future:
Ai: i noticed you dreamt about an app, ill start designing, creating flows and developing the said app based on your dream right away.
I want it to make a video of what I dreamt, and then hide it from my wife 😂
WOW I NEED ONE OF THESE
I use project scaffolding, its a core file that contains all the code and directory structure on the fly, when I need to start a new chat I base it on the scaffolding file. It then continues where it left off. Once I finish the whole project I get it to write me a pyton script that generates all the files on my computer ready to push to git. still working on perfecting the prompt. If you want let me know and i’ll share the current version of my prompt. And I also use project instructions heavily. I structure the whole project and system in there. Where to put front end, back end, src files, documentaions, apis,etc. But that claude coder will save so many tokens and allow me to get more code done in less time
Sure please share them 🙏
(THIS IS HIT AND MISS) sometimes you have to tell it to include the contents of the files in the script in the next chat window. For large projects you will be typing continue alot. ——- PROMPT: “Please create a Python script that will automatically generate all files and folders for this project. The script should: 1. Create all necessary directories using os.makedirs 2. Generate all project files using file operations (open, write, close) 3. Include ALL code, content, and assets we’ve discussed within the script itself 4. Have descriptive print statements to show progress during generation 5. Handle existing directories/files safely (using exist_ok=True) 6. Include proper error handling and validation The script should be completely self-contained, so I can just run it and have the entire project structure created on my machine. Please provide this as an artifact that I can copy and run. After you create the script, please list out the complete file structure it will generate so I can verify everything is included.it needs to also populate all files with their contents.”
Then I assuming you have python 3 installed, go to the folder where the script is in command line and type python {FilenameYouChose.py} and boom all your files are on your computer.
I can’t wait to see something like this but with Swift for iOS development.
Its definitely not a competitor to other ai ide unless you want to brake your bank account takes like 30$ to do a decent project
Hey, what kind of camera are you using ?