News for hackers. Stuff that matters.
Re:Qwen4-Max by dan_thinks on Sunday November 15, 2026 @09:28PM (#50513296)
This is exciting. The continued focus on efficiency alongside quality is encouraging - it makes me hopeful to see what smaller models will become capable of.

Looking forward to the "Flash" and 27B versions.

Re:Qwen4-Max by teaearlgraycold on Sunday November 15, 2026 @11:52PM (#50514176)
I'm also hoping someone puts out a competitive 30-70B model soon. There's this massive dead zone between 24GB and <100GB of VRAM where it seems your best option is Qwen 3.8 27B.
Re:Qwen4-Max by shockembopper on Sunday November 15, 2026 @11:58PM (#50514214)
There are many engines that alleviate this issue in particular. The current mood is running qwen flash next using strata, basalt, ect.
Re:Qwen4-Max by AtHeartEngineer on Monday November 16, 2026 @08:15AM (#50517286)
radiance
Re:Qwen4-Max by q17 on Monday November 16, 2026 @03:49AM (#50515464)
Same, I recently found Nemotron 3.5 74B-A9B really good at filling this gap though, could do (very slow) local inference on just 32GB DDR4, without the quality dropping to the floor
Re:Qwen4-Max by dstu on Monday November 16, 2026 @06:33AM (#50516421)
[delayed]
Re:Qwen4-Max by samtheprogram on Monday November 16, 2026 @09:59AM (#50518297)
[delayed]
Re:Qwen4-Max by seanmcdirmid on Monday November 16, 2026 @11:55AM (#50519766)
Qwen 3.8 27B is a dense model, so prefill times kind of suck on Mac’s. MoE models offer way more value unless yiu have the bandwidth to handle a dense model.
Re:Qwen4-Max by neurophant on Monday November 16, 2026 @10:47AM (#50518865)
Waiting for the Flash tier too. My experience running coding agents on cheap vs big models: the cheap one gets the same work done, roughly twice as slow. The big model buys you speed, not results.