Best LOCAL AI for programming

Discussion & Solutions

AI-generated summary

:bullseye: What you get: Practical advice for selecting and evaluating local programming AI models on mid-range hardware.

:white_check_mark: Start here:

  • Benchmark with a single, small coding task like refactoring one function or generating unit tests.
  • Monitor actual VRAM and RAM consumption closely while running the task.
  • Verify that generated code compiles and passes tests without manual fixing.

:warning: Watch out: Larger models that spill into system RAM often run slowly and produce verbose, unhelpful output instead of accurate code.

:brain: Takeaway: Keep a working model as a fallback and test candidate models side-by-side using identical prompts.

1

Hello everyone!

I am looking for the best uncensored local AI for programming. My setup is: Ryzen 9 7900, RTX 5060ti 16GB, 32GB RAM 5000mhz. I am currently running Ornith1.5 35B a3b q5_k_m and the speed is fine, it does not need to be fast it can spill into RAM. I am wondering isthis the best censored model I can run or there is a bigger or a smarter one or both. Thanks in advance!

2

With your 16GB card, I would start by testing any new model on one small, real coding task first, such as refactoring a single function or writing unit tests, before switching your whole workflow to it. Check how much VRAM and RAM it actually uses during that test, and whether the output compiles and passes your existing tests without extra fixing. A larger model is not always better if it is slow to load or keeps producing long, vague answers for simple edits. Keep your current setup working as a fallback, and compare the new model on the same prompt so the difference is clear. If it is stable and saves you real time on that small task, then try it on a larger file next.