Qwen3.8-Flash-Next · Q4_K_M
This page aggregates 1 real-world runs of Qwen3.8-Flash-Next (Q4_K_M) on NVIDIA RTX 5090 with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.
12.8 tok/s
Decode
Decode speed
96 tok/s
Prefill
Prefill speed
1.04 s
TTFT
Time to first token
32 GB
VRAM
VRAM usage
1
Measured runs
1
Independent sources
Other
Source platforms
Today
Last verified
Performance
- Strata · Q4_K_M Decode 16.4 · Prefill 123 ·
- Strata · IQ3_XXS Decode 28.4 · Prefill 213 ·
- Strata · IQ3_S Decode 24.6 · Prefill 184.5 ·
- Strata · IQ4_XS Decode 18.4 · Prefill 138 ·
- llama.cpp · Q4_K_M (current)Decode 12.8 · Prefill 96 ·
- llama.cpp · IQ3_XXS Decode 22.1 · Prefill 165.8 ·
- llama.cpp · IQ3_S Decode 19.2 · Prefill 144 ·
- llama.cpp · IQ4_XS Decode 14.4 · Prefill 108 ·
- vLLM · IQ3_S Decode 20.9 · Prefill 156.8 ·
Core figures
- Decode (avg)
- 12.8 tok/s
- Prefill (avg)
- 96 tok/s
- TTFT (avg)
- 1.04 s
- VRAM (avg)
- 32 GB
- MTP acceptance rate
- —
- TTFB
- — GB
- Power draw
- — W
Configuration
Member-level fields are taken from the most recent run
- Model
- Qwen3.8-Flash-Next
- Quantization
- Q4_K_M
- Framework
- llama.cpp
- Version
- b4200
- Context length
- 4096 tokens
- Batch size
- 512
- GPU layers
- 99
- Flash Attention
- On
Hardware
Nominal and measured figures are shown side by side; whether it runs is the reader's call
- GPU
- NVIDIA RTX 5090
- Nominal VRAM
- 32 GB
- Measured VRAM (avg)
- 32 GB
- OS
- Linux
- Driver
- 570.65
- CUDA
- 12.6
Get started
Original model resources
Base model resources may not match this quantization. Use the verified deployment weights above when available.
Sources & evidence
1 measured records in total, each traceable to its original source
b4200 · Linux · CUDA 12.6 · 4096 ctx
12.8 tok/s
Decode
96 tok/s
Prefill
1.04 s
TTFT
32 GB
VRAM
—
MTP
— W
Power draw
L0 理论估算（add-1000-records-batch3）