Verdict: Qwen3.8 much better at tool calling and general agentic work than 3.6, but still not a great coder DeepSeek still crushes it at coding

Achievement Unlocked: First quantized model and it actually fucking works https://huggingface.co/barrydeen/Qwen3.8-27B-AWQ-4bit

Qwen3.8 27b running locally on 2x 3090s Making it to one shot an outbox nostr client, let's see what it can do

More to come, run local

Nobody making an int4 quant for 3.8 27b so I guess I will have to do it

Doesn't fit on 2x 3090 with any good context

Can someone send me $8,000 I really need two more dgx sparks PLEASE, I know there is a billionaire whale out there somewhere

Mass censorship of bots is happening on NostrArchives - about 30% done so far Been running for days, identifying and deleting events Bots outnumber humans are least 100:1 on nostr :/

THE AI PSYCHOSIS WILL CONTINUE LET ME IN!!!!

Brave not to update, brave if we do

I used to hate china due to communism, but No Epstein pedos No wars And most importantly, free llms Thank you china

API key + prime-agent

Should be fixed this week, was a bit distracted last week 😔

Bcashjr

Kratter is new to bitcoin, his first stint got all his followers to buy the 2021 top, even though he had never seen a cycle before Then he moved on to getting everyone to follow a fork even though he was never around for the blocksize wars The lesson here is really to just stay humble

People on twitter will invest millions into bitcoin but suggesting they spend 2 minutes rolling dice is a bridge too far and proof that bitcoin ngmi apparently Like bros probably play monopoly for hours but won't do the same to keep their money safe smh

Toxicity is part of bitcoin culture gotta be good sports and shake hands and get back to destroying fiat

I wanna start having it do sensitive stuff for me don't wanna send all this shit to the cloud, like doing my accounting and stuff

Like 8kish

Gm, Alright, this is getting nutty Glm 5.3 is beating fable and Kimi k3 on multiple benchmarks, and it can run locally on 4 gb10s 🥵

If you're interested in the speed you can get using DeepSeek locally for about $7k, here it is I think for devs this is an absolute no brainer it actually does pay for itself vs Claude max, or multiple subscriptions