← Overview
qwen3.8-flash-next-ple4
- Quantization
- W4A16
- Reasoning
- off
- Profile
- quick
- Suite
- v2b
- Date
- 2026-09-17 – 2026-09-18
- Hardware
- NVIDIA DGX Spark, 128 GB RAM
- Inference software
- vLLM 0.26.0
- Measured
- directly at the inference server
- Overall
- 71.5
Categories
| Category | Score |
|---|
| Knowledge & reasoning | 77.9 |
| Math | 73.2 |
| Instruction following | 49.6 |
| Code | 75.5 |
| German | 81.1 |
Tasks
| Category | Task | Score | ± | n |
|---|
| Knowledge & reasoning |
mmlu_pro |
86.3 |
± 2.0 |
300 |
| Knowledge & reasoning |
gpqa_diamond_gen |
77.3 |
± 3.0 |
198 |
| Knowledge & reasoning |
musr_gen |
70.0 |
± 2.9 |
250 |
| Math |
omni_math_gen |
73.2 |
± 2.8 |
250 |
| Instruction following |
ifbench |
36.0 |
± 2.8 |
300 |
| Instruction following |
ifeval |
90.7 |
± 2.4 |
150 |
| Code |
livecodebench_gen |
75.5 |
± 3.0 |
200 |
| German |
mmlu_prox_de |
85.7 |
± 2.0 |
300 |
| German |
include_de_gen |
72.1 |
± 3.6 |
154 |
± is the standard error; the 95% interval spans roughly 1.96 standard errors around the score.
Speed
| Measure | Context (target) | Context (actual) | Median | p10 | p90 | n |
|---|
| Generation [t/s] |
– |
– |
26.4 |
26.3 |
26.6 |
5 |
| Prompt processing [t/s] |
512 |
529 |
1051.1 |
1048.4 |
1056.2 |
5 |
| Prompt processing [t/s] |
4096 |
4029 |
2041.5 |
2032.9 |
2044.5 |
5 |
| Prompt processing [t/s] |
16384 |
16012 |
2266.7 |
2263.1 |
2267.9 |
5 |
| Time to first token [ms] |
512 |
529 |
503.3 |
500.3 |
504.6 |
5 |
| Time to first token [ms] |
4096 |
4029 |
1973.6 |
1970.7 |
1982.0 |
5 |
| Time to first token [ms] |
16384 |
16012 |
7064.0 |
7060.4 |
7075.3 |
5 |
Measured directly at the inference server.