GPT 4o vs. Sonnet 3.5 for Coding?

[Next Press Pro]

For all the BHW programmers out there..

What's your take on GPT 4o v. Sonnet 3.5 for programming task?

Personally, I have found the code produced by Claude Opus 3.0 to be superior than ChatGPT. After several weeks of failing to resolve a programming issue with ChatGPT, I was able to solve it in less than one hour with Opus 3.0. Honestly, I'm blown away by Anthropic.

What's your take on OpenAI vs. Anthropic? Do you think Claude Opus 3.5 will rival GPT-5?

How has AI changed your coding workflow?

 

[SirSmash]

Neither. Run a local LLM specifically trained for programming like DeepSeek-Coder-V2, Codestral, or CodeLlama. That'll get you better results than either, especially if you can run the larger models (above 7b) that are less prone to hallucinating. You'll need a lot of VRAM though (or you got to wait longer with a CPU). The poor mans way is pickup an M40 or P40 and swap the cooler from a GTX 770 then you can run it in a regular desktop. 24GB VRAM lets you offload most LLM's to the GPU, with only the largest hitting system RAM too.

OpenAI vs Anthropic? OpenAI is going to win, they're better at marketing. It doesn't matter which product is better, the best product often doesn't win. Most people don't know anything but ChatGPT even exists.

 

[willywhiskers]

Neither. Run a local LLM specifically trained for programming like DeepSeek-Coder-V2, Codestral, or CodeLlama. That'll get you better results than either, especially if you can run the larger models (above 7b) that are less prone to hallucinating. You'll need a lot of VRAM though (or you got to wait longer with a CPU). The poor mans way is pickup an M40 or P40 and swap the cooler from a GTX 770 then you can run it in a regular desktop. 24GB VRAM lets you offload most LLM's to the GPU, with only the largest hitting system RAM too.

OpenAI vs Anthropic? OpenAI is going to win, they're better at marketing. It doesn't matter which product is better, the best product often doesn't win. Most people don't know anything but ChatGPT even exists.

What are the prerequisites to run a local LLM on our computers?

What level of python do I have to know ?

 

[neslex]

Sonnet 3.5 easy.

 

[SirSmash]

What are the prerequisites to run a local LLM on our computers?

What level of python do I have to know ?

You don't need to know any python.

 

[willywhiskers]

You don't need to know any python.

a little bit elaboration?

 

[provocative]

Claude 3.5 Sonnet

 

[robinwood]

As mentioned above by @SirSmash you should try deepseek. It's pretty good.

 

[Larry Igna]

You don't need any knowledge. Just get an appropriate llamafile and execute it.
This one for example launches a server with deepseekcoder
https://huggingface.co/rabil/deepseek-coder-6.7b-instruct-llamafile

 

[Yasua]

Which platform is more suitable for you depends on your usage scenario and needs. While GPT-4o offers a broad perspective for general programming and problem-solving, Sonnet 3.5 may be preferred for concentrating on a specific programming language and getting direct assistance when writing code.

 

[SirSmash]

a little bit elaboration?

Elaborate what? You asked if you need to know python, I said you don't need to know python. There's nothing to elaborate.

If you're looking for needed system specs, Google it perhaps? I don't even know if you're on a PC or Mac or what you're trying to accomplish here. Running 7b on a potato or 33b on a Tesla GPU? You got to do some research; I'm not going to handhold you step-by-step. It's the internet, just like you don't care if I can do a specific task/skill, I don't care if you can run a local LLM. That's just the honest truth. I got nothing invested in your ability, or inability, to pull it off.

 

[rafark]

I’ve given Claude more than one chance but I always find myself going back to chatgpt.

 

[Next Press Pro]

Neither. Run a local LLM specifically trained for programming like DeepSeek-Coder-V2, Codestral, or CodeLlama. That'll get you better results than either, especially if you can run the larger models (above 7b) that are less prone to hallucinating. You'll need a lot of VRAM though (or you got to wait longer with a CPU). The poor mans way is pickup an M40 or P40 and swap the cooler from a GTX 770 then you can run it in a regular desktop. 24GB VRAM lets you offload most LLM's to the GPU, with only the largest hitting system RAM too.

OpenAI vs Anthropic? OpenAI is going to win, they're better at marketing. It doesn't matter which product is better, the best product often doesn't win. Most people don't know anything but ChatGPT even exists.

I agree that OpenAI has First-mover advantage. But if Anthropic continues to push out new features and new models like it is doing, it can take over the market (it won't be easy but it's possible)

I think local LLM is OK when the task is not intricate and does not need a high level of accuracy. For complicated coding tasks I prefer using one of the closed source models.

 

[Next Press Pro]

What are the prerequisites to run a local LLM on our computers?

What level of python do I have to know ?

You don't need to know python to run local LLM. Python is a programming language it is not related to LLM.

It is helpful to have basic coding knowledge but there are several tools that can help you if you are new to this:

• LM Studio.ai: Run local LLM using graphical interface
• GPT4ALL
• Ollama: This tool require command line knowledge

I hope this helps!

 

[Next Press Pro]

I’ve given Claude more than one chance but I always find myself going back to chatgpt.

Interesting. Can you elaborate what specific issues you encountered with Claude? Which model did you test?

 

[Next Press Pro]

As mentioned above by @SirSmash you should try deepseek. It's pretty good.

I've heard some great things about DeepSeek-Coder-V2 but never used it personally (it ranks pretty high in lmsys.org leadboard)

 

[robinwood]

I've heard some great things about DeepSeek-Coder-V2 but never used it personally (it ranks pretty high in lmsys.org leadboard)

try it ... its free.

 

[Hannah Ferguson]

In summary, GPT-4o and Claude 3.5 Sonnet differ significantly in terms of speed, performance, functionality, cost and usability. GPT-4o demonstrates excellent performance in several domains with its outstanding real-time reasoning capability and multimodal version; while Claude 3.5 Sonnet is more comprehensive in terms of speed, performance and functionality, with particular advantages in encoding capability and complex task processing.

 

[rafark]

Interesting. Can you elaborate what specific issues you encountered with Claude? Which model did you test?

The latest model. Honestly I don’t remember but probably something related to react & typescript which is what I’m working with right now. Chatgpt always solves my problem at the first try. When I tried to have Claude help me it just couldn’t.

 

[Next Press Pro]

The latest model. Honestly I don’t remember but probably something related to react & typescript which is what I’m working with right now. Chatgpt always solves my problem at the first try. When I tried to have Claude help me it just couldn’t.

Aha. I see. I haven't used react or typescript so maybe therein lies the difference.