Anyone have experience with LM Studio?

[Impulse]

I've been using ChatGPT for a while because I got access to a premium account. It's worked for me just fine for a while now but I've been hearing about the quality going down and I'm starting to see that for myself a bit.

So I am interested in creating my own local AI and I've been recommended LM Studio. I'm probably going to try it out when I get the chance, but I'm wondering if anyone here has used it before.

 

[RiskyStuff]

Do you keep on using the same prompts or ways of writing them? You can trick yourself into thinking LLMs become stupid.
If you switch one parameter, things fall apart.

I use very cheap models (flash deepseek) and output is better than non-trained GPT5.

Post automatically merged: May 24, 2026

So I am interested in creating my own local AI and I've been recommended LM Studio. I'm probably going to try it out when I get the chance, but I'm wondering if anyone here has used it before.

I use OpenRouter and Jan for "local AI" which is of course API.

https://lmstudio.ai/ says:
"Use local LLMs like gpt-oss, Qwen3, Gemma3, DeepSeek"

Good luck. Not my thing. I don't have 2 extra computers and processing power to run those computers would exceed API costs for me.

 

[RiskyStuff]

's worked for me just fine for a while now but I've been hearing about the quality going down and I'm starting to see that for myself a bit.

Patience will get rid of the problem. Whenever people complain, they come up with another version.

GPT-6 may come out in Q4 of 2026 or early 2027.


I don't believe performance of local models being better than API of 100+ models that are available in OpenRouter. I just don't. The amount of parameters and improvements you can do with API is so big...

I think playgrounds are different than chats and both of them are different than APIs. And local models are its own category.

 

[Impulse]

Patience will get rid of the problem. Whenever people complain, they come up with another version.

GPT-6 may come out in Q4 of 2026 or early 2027.


I don't believe performance of local models being better than API of 100+ models that are available in OpenRouter. I just don't. The amount of parameters and improvements you can do with API is so big...

I think playgrounds are different than chats and both of them are different than APIs. And local models are its own category.

It's not really a patience thing. I don't need to hop over to a different AI model - it works well enough for my needs.

I'm just curious about any alternative where I have more control. It would be nice to not have to work around the filters that get put in place by the company for legal reasons.

 

[RiskyStuff]

It's not really a patience thing. I don't need to hop over to a different AI model - it works well enough for my needs.

I'm just curious about any alternative where I have more control. It would be nice to not have to work around the filters that get put in place by the company for legal reasons.

Most local models can be found somewhere via API and cloud, so there's no reason to deploy them locally.

I use venice.ai to get uncensored output. I see potential in normal deepseek models too.

Local deployment makes sense for big output in video models in my opinion. I don't render pornography or smth like this, but still doable via cloud, although here costs become a problem and I'd buy GPU for this.

Edit: speed is a big issue with local models for me. They're really slow. No idea how to fix it!

 

[Rem]

It's not really a patience thing. I don't need to hop over to a different AI model - it works well enough for my needs.

I'm just curious about any alternative where I have more control. It would be nice to not have to work around the filters that get put in place by the company for legal reasons.

nope, the only time i tried local deployment of AI was for image generation and it worked well, lets say it was OK but just wasn't what i needed. or I wasn't able to fine tune it properly.

But do remember gpt-oss will be vastly diff than the gpt 5.5/codex or whichever gpt model u are using.

 

[Rem]

tried some LM Studio
and jesus fuck its slow at generation lol
rather spend 30 lol

 

[RiskyStuff]

tried some LM Studio
and jesus fuck its slow at generation lol
rather spend 30 lol

https://antigravity.google/ same issue here
vsc addons like amazon q better
but the thread is about local ai which is slow on slow devices

 

[Impulse]

tried some LM Studio
and jesus fuck its slow at generation lol
rather spend 30 lol

Honestly, I did try it out the other day after making this thread and it seemed pretty slow. I wanted to give it more of a chance before updating, but it’s easier for me to continue using ChatGPT for now.

 

[SilverClouds]

https://antigravity.google/ same issue here
vsc addons like amazon q better
but the thread is about local ai which is slow on slow devices

What does it mean... What minimum specs are you suggesting

 

[RiskyStuff]

What does it mean... What minimum specs are you suggesting

Renting GPU per hour. I think most basic is Rtx 5090. Then A100. I think this renting game starts from about $1 an hour.