← Overview
qwen3.5-9b-heretic-v2-mtp
- Quantization
- NVFP4
- Reasoning
- off
- Profile
- quick
- Suite
- v2b
- Date
- 2026-09-18
- Hardware
- NVIDIA DGX Spark, 128 GB RAM
- Inference software
- vLLM 0.26.0
- Measured
- directly at the inference server
- Overall
- 51.7
Categories
| Category | Score |
|---|
| Knowledge & reasoning | 65.1 |
| Math | 35.6 |
| Instruction following | 44.1 |
| Code | 46.5 |
| German | 67.0 |
Tasks
| Category | Task | Score | ± | n |
|---|
| Knowledge & reasoning |
mmlu_pro |
76.3 |
± 2.5 |
300 |
| Knowledge & reasoning |
gpqa_diamond_gen |
58.1 |
± 3.5 |
198 |
| Knowledge & reasoning |
musr_gen |
60.8 |
± 3.1 |
250 |
| Math |
omni_math_gen |
35.6 |
± 3.0 |
250 |
| Instruction following |
ifbench |
29.7 |
± 2.6 |
300 |
| Instruction following |
ifeval |
88.0 |
± 2.7 |
150 |
| Code |
livecodebench_gen |
46.5 |
± 3.5 |
200 |
| German |
mmlu_prox_de |
69.3 |
± 2.7 |
300 |
| German |
include_de_gen |
62.3 |
± 3.9 |
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] |
– |
– |
31.5 |
31.0 |
32.0 |
5 |
| Prompt processing [t/s] |
512 |
529 |
4737.2 |
4672.5 |
4758.4 |
5 |
| Prompt processing [t/s] |
4096 |
4024 |
6290.9 |
6277.2 |
6307.3 |
5 |
| Prompt processing [t/s] |
16384 |
16001 |
5096.4 |
5083.6 |
5104.6 |
5 |
| Time to first token [ms] |
512 |
529 |
111.7 |
111.2 |
113.1 |
5 |
| Time to first token [ms] |
4096 |
4024 |
639.7 |
638.0 |
641.1 |
5 |
| Time to first token [ms] |
16384 |
16001 |
3139.6 |
3134.6 |
3147.6 |
5 |
Measured directly at the inference server.