Qwen3.8-Flash-Next (tp=2) from Nvidia (FP4-Local)
Performance is up ~20%-30% comparing to TP=1.
🧠 KV - 4.18m ‼️+ YaRN
🚀 Peak c=1 @ 92 tok/s
🏭 Peak c=64 @ 834 tok/s
Latest 0.32.0 vLLM
⚡️ github.com/myllmbox/qwen38-f…
Qwen3.8-Flash-Next from Nvidia (FP4-Local)
Same amazing speed and better quality‼️
🧠 KV - 1.66m ‼️+ YaRN
🚀 Peak c=1 @ 82 tok/s
🏭 Peak c=16 @ 297 tok/s
Latest 0.32.0 vLLM
⚡️ github.com/myllmbox/qwen38-f…
5
1
17
891
Full comparison of TP=1 vs TP=2 could be found here myllmbox.com/?a=mbx-v53&b=mb…
76
Latest Nvidia FP4-Local quants are just amazing 🥰
So what to chose?
Qwen3.8-Flash-Next or GLM-5.3-Flash
We compared them so you could take a look at the differences and make your choice 😉
⚡️ myllmbox.com/?a=mbx-v53&b=mb…
Qwen3.8-Flash-Next from Nvidia (FP4-Local)
Same amazing speed and better quality‼️
🧠 KV - 1.66m ‼️+ YaRN
🚀 Peak c=1 @ 82 tok/s
🏭 Peak c=16 @ 297 tok/s
Latest 0.32.0 vLLM
⚡️ github.com/myllmbox/qwen38-f…
3
1
23
2,097
Never listen to French songs, but this AI "slop" is amazing ‼️
I came to realize people liked the french #slopcore without having the lyrics 😅 sorry!
So here is a version with english subtitles!
1
4
328
Qwen3.8-Flash-Next from Nvidia (FP4-Local)
Same amazing speed and better quality‼️
🧠 KV - 1.66m ‼️+ YaRN
🚀 Peak c=1 @ 82 tok/s
🏭 Peak c=16 @ 297 tok/s
Latest 0.32.0 vLLM
⚡️ github.com/myllmbox/qwen38-f…
7
10
92
7,914
@AgentSparko @yume_arasaki i guess you guys will try this one? 🤣 all other stats at myllmbox.com/?a=mbx-v53&b=mb…
1
2
536
GLM-5.3-Flash from Nvidia (FP4-Local)
🚀 Peak c=1 @ 69 tok/s
📊 Average c=1 @ 42 tok/s
Most important thing quality ‼️
In our visual tests it scored 11.5 out of 12 ‼️
github.com/myllmbox/glm53-fl…
6
4
64
4,940
myllmbox.com/?a=mbx-v2-tp2-g… to see how it performs vs v1. V1 was bit quantized on the fly this one is fully original. @aldozampatti made a nice quant according to v1 that runs ~20% faster and scales good at bigger c. That one will land soon too
1
1
4
458
Qwen-Image-2.1-Turbo on one 16GB card 🤯
Yes you read it right 5080 can easily run it at amazing speed.
Generate
🖼️1 MP -> 6.3s
🖼️2 MP (hd 16x9) -> 12s
Edit
🖼️ 1 MP + 🖌️ -> 8.5s
🖼️ 2 MP (hd 16x9) + 🖌️ -> 14.8s
⚡️ github.com/myllmbox/qwen-ima…
3
1
5
683
What was longest session you had opus working no stooping and asking some 🤪? 🤣
Qwen3.8-Flash-Next is already on it's way without interruptions orchestrating cycle for close to 8h.
8h - with no stupid questions ‼️🤯
Qwen3.8-Flash-Next tp=2‼️
DS4.1-Flash looked like a miracle running at 200+ now we have similar speed for Qwen3.8 locally 177tok/s‼️
🚀 Peak c=1 @ 177 tok/s
📊 Average c=1 @ 93 tok/s
🏭 Peak c=64 @ 1,174 tok/s
⚙️ Prefill 3,939 tok/s
💻 TTFT 0.36s
⚡️ github.com/myllmbox/qwen38-f…
9
2
11
1,123
@grok doing this till sub maxed then @Alibaba_Qwen took over 🤣. Qwen was doing other tests during that time 😆
1
186
That's how you need to build things❗️
Only then you can expect mainstream using it 🥰
TensorFold now has a "Beta" macOS app to launch the engine, so you don't need to jump into the CLI.
It's also under /app in the git repo if you want to drop any PR's against it.
tensorfold.dev/
7
496
LFM2.5-24B-A2B by @Blackfrost_AI
For those who need some fire power on 5080 and similar hardware with just 16GB. 😉
🚀 Peak c=1 @ 225 tok/s
🏭 Peak c=4 @ 571 tok/s
⚙️ Prefill 17,157 tok/s
💻 TTFT 0.15s
⚡️ github.com/myllmbox/lfm25-24…
2
8
600
vLLM 🆚 TensorFold
Same Qwen3.8-Flash-Next quant by azampatti. Different engines - different results ‼️
TensorFold - leads on c=1 structured / code and sometimes a lot ‼️Peak 238.7 vs 112.8 🤯
vLLM - leads in concurrency 🏆
⚡️ myllmbox.com/?a=mbx-v52&b=te…
Qwen3.8-Flash-Next in 4 days got ~30% boost‼️
We finally have pushed it above 100 on tp=1 🥳🎉
🚀 Peak c=1 @ 111 tok/s
📊 Average c=1 @ 62 tok/s
🏭 Peak c=16 @ 414 tok/s
⚙️ Prefill 2,727 tok/s
💻 TTFT 0.57s
⚡️ github.com/myllmbox/qwen38-f…
10
1
30
3,147