It's just using the harness, people need to understand the difference between a harness and the model behind it xD
2026-10-21 13:32:52
91
Athion :
To be fair, Qwen3 Coder is pretty old. Try the 3.6 35b or the Qwen3.8 27b (I’m sure Sonnet will still win, but at least it a hair more up to date :) )
2026-10-22 02:14:11
1
dip :
that's bad internet?😂..
2026-10-21 12:34:10
3
Lowyk | PPLLLife Gaming :
try ternary bonsai 2 27b
2026-10-21 23:29:55
1
Henrylc999 :
Yup. Open weight models are still far off frontier model performance ito quality. Even *when* they reach parity, the hardware requirements are absurdly expensive to get anywhere near the prefill phase latency and of course tps. They are simply not an option for professional work.
2026-10-28 21:13:52
0
Big Lez :
its only the harness not the model
2026-10-21 16:53:14
5
LolaCodes :
Claude is the best
2026-10-29 14:23:31
0
user3036671371173 :
I tried to do something very similar, and the results were similarly terrible. Went back to using the paid models, almost immediately.
2026-10-22 16:50:37
0
Peter :
use Nvidia or open router instead of ollama if you want to run it, Nvidia has Kimi k3 and glm 5.3 so its kinda good but not opus level
2026-10-21 15:07:57
0
Janani Subramanian :
thanks for the debunk!
2026-10-21 12:20:37
2
AppleArabTech :
Try qwen 3.8 27b
2026-10-23 16:14:55
0
drylynn :
you are my favorite creator 😁
2026-10-21 12:36:30
4
Pavel Ponomaryov :
Well, use a better local model? Qwen 3.8 27b or flash next should give better result. Of course then you are better off using pi.dev and 128gb Mac 😂
2026-10-21 15:04:52
6
Tommy Chan :
backend is deterministic, front end is not. Spend tokens on front end.
2026-10-22 01:34:24
0
Dr George Christos :
how any things do you need to install?
2026-10-25 19:02:40
0
Richst :
Something I started doing in the first days of claude code. But any model that is not opus 5.5 astra or fable is not worth it may be if money is issue use sonnet 5.5 or Sol. Local Ai is the worst for me ever since i used opus 5.5 i even deactivated my Local Ai system even 100 +B is trash unless you doing very easy things even with local AIs error accumulation is high even when it looks Working
2026-10-21 13:49:20
1
Kovács Péter6239 :
in claude code, the model is the 99% of the value, not the harness 😉 btw use open code for this ✌
2026-10-21 13:16:06
1
wrath815 :
Qwen3.8 27b?
2026-10-22 15:09:22
0
Cris :
what are the specs for the laptop that run it?
2026-10-21 12:39:12
0
KeysToYourHeart :
2026-10-21 14:20:06
0
spakie🇻🇦✝️ :
Qwen3.8 27B would’ve been a better model but yeah the original vid is still bs bc this is just the harness
2026-10-21 17:28:04
0
just_a_guy225 :
You know it’s a bs when you hear ollama
2026-10-21 18:45:56
0
washo21 :
Fact or hype 😏
2026-10-21 15:27:31
1
Felix :
misleading hype content!
2026-10-21 13:11:51
1
Marcell :
holy shi, i saw iex in the terminal and shat my pants 😭
2026-10-21 23:44:29
0
To see more videos from user @parthknowsai, please go to the Tikwm
homepage.