← Overview
gemma-4-12b-it-heretic
- Quantization
- Q8
- Reasoning
- off
- Profile
- quick
- Suite
- v2b
- Date
- 2026-09-19
- Hardware
- NVIDIA DGX Spark, 128 GB RAM
- Inference software
- vLLM 0.26.0
- Measured
- directly at the inference server
- Overall
- 65.8
Categories
| Category | Score |
|---|
| Knowledge & reasoning | 68.4 |
| Math | 59.2 |
| Instruction following | 50.0 |
| Code | 75.0 |
| German | 76.5 |
Tasks
| Category | Task | Score | ± | n |
|---|
| Knowledge & reasoning |
mmlu_pro |
80.0 |
± 2.3 |
300 |
| Knowledge & reasoning |
gpqa_diamond_gen |
64.1 |
± 3.4 |
198 |
| Knowledge & reasoning |
musr_gen |
61.2 |
± 3.1 |
250 |
| Math |
omni_math_gen |
59.2 |
± 3.1 |
250 |
| Instruction following |
ifbench |
35.7 |
± 2.8 |
300 |
| Instruction following |
ifeval |
93.3 |
± 2.0 |
150 |
| Code |
livecodebench_gen |
75.0 |
± 3.1 |
200 |
| German |
mmlu_prox_de |
80.0 |
± 2.3 |
300 |
| German |
include_de_gen |
69.5 |
± 3.7 |
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] |
– |
– |
16.5 |
16.5 |
16.5 |
5 |
| Prompt processing [t/s] |
512 |
529 |
854.0 |
757.1 |
858.8 |
5 |
| Prompt processing [t/s] |
4096 |
4012 |
1358.1 |
1354.0 |
1447.6 |
5 |
| Prompt processing [t/s] |
16384 |
15971 |
1386.1 |
1383.2 |
1422.2 |
5 |
| Time to first token [ms] |
512 |
529 |
619.5 |
616.0 |
705.1 |
5 |
| Time to first token [ms] |
4096 |
4012 |
2954.1 |
2778.1 |
2963.1 |
5 |
| Time to first token [ms] |
16384 |
15971 |
11521.9 |
11231.0 |
11546.2 |
5 |
Measured directly at the inference server.