4× RTX 3090
110tok/sEveryday
1 request, prose prompt, with MTP (per recipe) Source
Intelligence40with thinking
- License unclear
Local AI · Recipe · 2× DGX Spark
Qwen3.8-Flash-Next NVFP4 with vLLM on 2× DGX Spark (text)
Qwen3.8-Flash-Next NVFP4 with vLLM on 2× DGX Spark (text): 121 tok/s according to github.com (dataset as of Sep 29, 2026).
by myllmbox
Engine
- Engine
- vLLM
- Quantization
- hibrid48: NVFP4-Output-Head, bf16-KV (41G/Box-Pin), halbe NVFP4-n-gram-Tabelle pro Box, Marlin MoE
- Model family
- Qwen3.8-Flash-Next
- Context
- 262,144
- Creator
- myllmbox
- GitHub stars
- 79
- Repo updated
- Sep 26, 2026
Measurements
Every sourced value of this recipe, each with its condition and source. Bars relative to the largest value in the group.
Decode · 1 request
What you need
- Hardware
- 2 × NVIDIA DGX Spark (GB10)
- Weights
- Main weightsmyllmbox/Qwen3.8-Flash-Next-hibrid48 other
- BaseQwen/Qwen3.8-Flash-Next other
- Engine
- vLLM
- Context
- 262,144 tokens
Notes
What matters before you rebuild it.
- Custom kernel required
- License unclear
Sources
Related recipes
2× DGX Spark
54.4tok/sEveryday
1 request, prose prompt, with MTP (per recipe) Source
Intelligence40with thinking
- Access on request
- License unclear
1× DGX Spark
53tok/sEveryday
1 request, prose prompt, with MTP (per recipe) Source
Intelligence40with thinking
- Custom kernel required
- License unclear
1× DGX Spark
48.7tok/sEveryday
1 request, prose prompt, with MTP (per recipe) Source
Intelligence40with thinking
- Access on request
- License unclear
2× DGX Spark
43.2tok/sEveryday
1 request, prose prompt, with MTP (per recipe) Source
Intelligence40with thinking
- Custom kernel required
- License unclear
2× RTX 3090
36.8tok/sEveryday
1 request, prose prompt, with ngram Source
Intelligence40with thinking
- Partly in system RAM
- Experimental
- License unclear
Local AI in your company?
In a workshop we work out which models and which hardware fit your tasks, and build the first agent on your infrastructure.