Hacker Newsnew | past | comments | ask | show | jobs | submitlogin
Ask HN: Best local LLM options under 2500 USD?
1 point by 999900000999 8 days ago | hide | past | favorite | 5 comments
Let’s say you have about 2500$ to spend on hardware, and want to start with local LLMs.

What would be the best deal?

 help



What are you using your local LLM for? What's the smallest/weakest model (and probably at what speed) that would make the hardware spend worth it?

having your own hardware lets you do much more than serve LLMs, it's a great learning platform in general

the other thing is that the little models are getting better and more capable, so the hardware ROI becomes greater over time


Ideally I want to even train or at least fine tune, but I can’t afford a Blackwell.

I'd recommend to start by using a platform for first experiments, like fine-tuning offerings, which can be done for lower cost and will handle the thornier issues therein.

Do some experiments to see if fine-tuning shows potential for improvements, curating the dataset and reward function are the harder parts anyway.


I’d like coding assistance and light logic.

Maybe the DeepSeek R1 distilled




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: