Qwen3.8-Flash-Next · UD-Q3_K_XL
This page aggregates 1 real-world runs of Qwen3.8-Flash-Next (UD-Q3_K_XL) on 4x NVIDIA Tesla P40 with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.
21.6 tok/s
Decode
Decode speed
330 tok/s
Prefill
Prefill speed
— s
TTFT
Time to first token
— GB
VRAM
VRAM usage
1
Measured runs
1
Independent sources
GitHub
Source platforms
Yesterday
Last verified
Performance
- llama.cpp · UD-Q3_K_XL (current)Decode 21.6 · Prefill 330 ·
Core figures
- Decode (avg)
- 21.6 tok/s
- Prefill (avg)
- 330 tok/s
- TTFT (avg)
- — s
- VRAM (avg)
- — 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
- UD-Q3_K_XL
- Framework
- llama.cpp
- Version
- b11058
Hardware
Nominal and measured figures are shown side by side; whether it runs is the reader's call
- GPU
- 4x NVIDIA Tesla P40
- Nominal VRAM
- 96 GB
- OS
- Linux
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
b11058 · Linux
21.6 tok/s
Decode
330 tok/s
Prefill
— s
TTFT
0 GB
VRAM
—
MTP
— W
Power draw
Qwen3.8-Flash-Next, no speculation | 20.6 | 20.0 | 21.6 ... Prefill: 122B about 460-477 t/s, Flash-Next about 320-330 t/s.