FRAMEWIREIndonesiaUpdated Aug 15Live wire
0:00 / 0:00

Testing Qwen 3.8 from Alibaba_Qwen with 3sparks chat on my iPhone with the model running on my Framework desktop in lmstudio.

I’m getting 19-25tks on the Framework's AMD AI MAX 395 while on another PC with an AMD 7900XTX with 24 gigs of GPU mem, I’m getting around 30 tks.

Sebastien BarrauAug 15
0:00 / 0:00

Qwen 3.8 27B is pretty decent so far, one shot this flappy bird in html rather quickly without bugs...

Daniel OuAug 159
0:00 / 0:00

Qwen 3.6-27B is similar to Qwen 3.8-27B

AkshAug 156
0:00 / 0:00

Cute little threejs game by Qwen 3.8 29B, Q8 unsloth running Hermes Agent...

Very short prompt with little specs and constraints, four five sentences maybe... It took a long time, since this model thinks a LOT... it needs more than 128k ctx in Hermes to avoid compaction-loop on

Andre Buckingham 🧙‍♂️Aug 15