Local AI is getting really really good.

This is the only news I care about these days!

the intelligence chart has like six models bunched within a few points, that's not a gap that's noise

im getting more and more hyped for local ai

Are you building this? Looks neat, we’d love your thoughts on which is also open-source and is already live

Love it!

Is GLM 5.2 considered "local AI" lol Definetely really want to try Step 3.7 Flash when I get my two sparks.

I want this to run on my MBP.

Running Glm-5.2 a 754 billion parameter model that takes 1.5 tb at BF16 at the same quality with the advancements in compression at 120 tok/sec is incredible.

Hangisi? (Donanım nedir bide)

Can I get access to please?

gimme

@ivanfioravanti The local AI was already really good.

Lol I was thinking some new model came out that made you say this. I forgot about the website.

im curious… if you had to give a ballpark estimate, what percent of present day AI usage could be done by local models? (And by local I mean small enough to run on consumer hardware. Not “open source”, but specifically local)

Depends on what you do for work I’d say most things can be done locally with 2-5k usd

Is vllm docker integration with local AI working now? For some reason my agent has trouble getting it to work a couple of weeks ago

Building sglang on dgx spark makes me want to yeet myself. It's retarded...

Absolutely, Local AI is already production ready.

Hi, are you happy with I’d like to avoid reinventing a benchmarking system for the hw/model comparisons I’m doing, and it looks pretty neat :-)

Model & Hardware?