Is GPT-4o better to use or is Claude 3.5 sonnet better to use?

Discussion
1

I am currently learning backend development, but I have a problem that is bothering me. I found that the two models in Cursor, GPT-4o and Claude 3.5 Sonnet, generate good quality code for my projects. If I need to write larger projects (what I consider large-scale) in the future, which model should I choose?

2

According to my experience, you can choose to

  • use o3-mini to design the architecture,
  • use Sonnet 3.5 to write the specific code,
  • use DeepSeek R1 to troubleshoot problems.

This is the best combination I think at present.

Reason:

  • The o3 - mini has a broader perspective and is suitable for designing the overall structure and process of the code.
  • Sonnet is more suitable for writing code as it can understand and follow your design.
  • DeepSeek is better at analyzing problems and is suitable for helping you solve difficult problems and propose solutions.
2 Likes
3

1000% this.

I’m in Agent mode when i want things done. Which is 90% of the time.
I use Claude 90% of the time.

I am rarely in “normal” mode, but when I am, i wish the LLM to brainstorm with me.
o3-mini and Deepseek have been equally surprisingly good at discussing architecture, networks, benchmarks, database optimizations, teaching me stuff, debugging. I can’t make a clear reason to use one over the other - i mostly use Deepseek purely becasue Cursor hosts their own so its cheap and more private.

4

Claude 3.5 Sonnet > GPT-4o

5

In my opinion, 4o still a bit lazy to provide full code. But if you use Claude, you usually see the response you want to see if you provide enough propmt.

I recommend o3-mini for planning and Claude for development just like the @tangjun and @markomitranic.