Local AI · Recipe · 2× DGX Spark

Qwen3.8-Flash-Next 125B NVFP4 with vLLM on 2× DGX Spark

Qwen3.8-Flash-Next 125B NVFP4 with vLLM on 2× DGX Spark: 54.4 tok/s according to github.com (dataset as of Nov 3, 2026).

by MiaAI-Lab

54.4tok/sEveryday

1 request, prose prompt, with MTP (per recipe)

Source: github.com

Intelligence (original model) · Artificial Analysis40with thinking

Instructions in the README Repo

Engine: vLLM vllm/vllm-openai:qwen38-flash-next
Quantization: NVFP4 (nvidia) + FP8 KV
Model family: Qwen3.8-Flash-Next
Context: 262,144
Parameters: 125B-A6B (s.o.)
Creator: MiaAI-Lab
GitHub stars: 384
Repo updated: Oct 4, 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

Everyday (decode)54.4 tok/sProse, TTFT 160 ms Source

Decode · parallel requests

Sums over all streams are marked; per request it is correspondingly less.

sum over 16 streams326 tok/sPeak (decode) · context 256K · Structured ×16 aggregate, vLLM-0.30-Lane (opt-in, 2026-09-27): 76,0 ×1 / 326,3 ×16; Code 313,6 ×8. Day-0-Lane: prose ×1 54,4, ×8 207,0 Source

Prefill by context

16K3,490 tok/s16k-Prompt in der v0.30-Lane (2026-09-27): 2.791 @1k / 3.313 @4k / 3.490 @16k / 2.982 @32k / 2.046 @64k / 2.898 @128k Source

What you need

Hardware: 2 × NVIDIA DGX Spark (GB10)
Weights: Main weightsnvidia/Qwen3.8-Flash-Next-NVFP4 other
otherRadixArk/Qwen3.8-Flash-Next-NVFP4 other
otherlocal-inference-lab/Qwen3.8-Flash-Next-NVFP4 other
BaseQwen/Qwen3.8-Flash-Next-FP8 other
Engine: vLLM vllm/vllm-openai:qwen38-flash-next
Context: 262,144 tokens

Sources

Related recipes

1× Mac

Qwen3.8-Flash-Next 125B-A6B MLX 4/8-bit with mlx-serve on 1× Mac (M5 Ultra)

by drowzeys

155tok/sEveryday

1 request, llmprobe standard prompt, thinking on, with MTP + PLD (per recipe) Source

Intelligence40with thinking

Experimental
License unclear

MLX mixed 4/8-bit (ddalcu, 100 GiB), KV bf16mlx-serve 26.9.7-dev @933566c, MLX 0.32.3 (NAX an)Weights Repo Updated Sep 30, 2026

Details

4× RTX 3090

Qwen3.8-Flash-Next 125B with llama.cpp on 4× RTX 3090

by tonyd2wild

110tok/sEveryday

1 request, prose prompt, with MTP (per recipe) Source

Intelligence40with thinking

License unclear

W4A16 (vLLM) bzw. GGUFvLLM (Lane 1) / llama.cpp (Lane 2)Weights Repo Updated Oct 1, 2026

Details

2× DGX Spark

Qwen3.8-Flash-Next 125B INT4-AutoRound with TensorFold on 2× DGX Spark (1M context)

by MiaAI-Lab

89.3tok/sEveryday

1 request, prose prompt, with MTP (per recipe) Source

Intelligenceno independent value

Custom kernel required
License unclear

INT4-AutoRound (GPTQ int4, Top-5-Experten) + FP8 n-gram-Tabelle, FP8 KVTensorFold tensorfold-native (Zig, zig-flashnext db281878 + patches), Image ghcr.io/miaai-lab/qwen3.8-flash-dual-dgx-sparks-tensorfold:zig-db28187-f3e1300f4f62, TP=2 über CX7/RoCE, MTP + Copy-Drafts (QUANT=int4ar)Weights Repo Updated Oct 26, 2026

Details

2× DGX Spark

Qwen3.8-Flash-Next INT4-AutoRound (v5.2) with vLLM on 2× DGX Spark

by myllmbox

74tok/sEveryday

1 request, prose prompt (long story), thinking off, with MTP Source

Intelligence40with thinking

Custom kernel required
License unclear

INT4-AutoRound (azampatti): int4-Experten, fp8-Seitenschichten; n-gram-Tabelle beim Laden fp8 → NVFP4, halb pro Box auf der GPU; KV-Pin 29G/Box (1.829.182 Tokens)vLLM 0.30 (Image myllmbox/qwen38-flash-next-cluster-vllm:v5.2; RecoverSSM, dynamischer Draft-Depth 3–7 in Full-CUDA-Graphs, sampled Drafts + Block-Verifikation, One-Shot-RoCE-All-Reduce, Dual-Rail, TP=2, 64 Seats) — Kit v5.2Weights Repo Updated Oct 29, 2026

Details

4× DGX Spark

Qwen3.8-Flash-Next NVFP4 with SparkRing on 4× DGX Spark

by fujitsupolycom

73tok/sEveryday

1 request, decode at 32K context (speed table, ring of four), thinking on, xhigh Source

Intelligence40with thinking

Custom kernel required

NVFP4 QAD (Local Inference Lab)vLLM (local-inference-lab Fork) via SparkRing (SIRCL-Collectives, switchless RDMA ueber ConnectX-7), Image Release 2026.10.2Weights Repo Updated Nov 2, 2026

Details

2× DGX Spark

Qwen3.8-Flash-Next MLX 4-bit with TensorFold on 2× DGX Spark

by sfxnz

69.6tok/sEveryday

1 request, prose prompt, with MTP (per recipe) Source

Intelligence40with thinking

License unclear

MLX affine 4-bit (group 32) auf allen Linears inkl. n-gram-Tabellen und MTP-Head, KV bf16TensorFold v0.6.2 (56e2e3e), TP=2, Image tf-qwen38-flashnext:0.6.2 (nvcr.io/nvidia/pytorch:26.07-py3)Weights Repo Updated Oct 2, 2026

Details

← Back to overview

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.

See the AI workshop Back to overview

Engine

Measurements

Decode · 1 request

Decode · parallel requests

Prefill by context

What you need

Sources

Related recipes

Qwen3.8-Flash-Next 125B-A6B MLX 4/8-bit with mlx-serve on 1× Mac (M5 Ultra)

Qwen3.8-Flash-Next 125B with llama.cpp on 4× RTX 3090

Qwen3.8-Flash-Next 125B INT4-AutoRound with TensorFold on 2× DGX Spark (1M context)

Qwen3.8-Flash-Next INT4-AutoRound (v5.2) with vLLM on 2× DGX Spark

Qwen3.8-Flash-Next NVFP4 with SparkRing on 4× DGX Spark

Qwen3.8-Flash-Next MLX 4-bit with TensorFold on 2× DGX Spark

Local AI in your company?