← Overview
mimo-v2.6-flash (pool)
- Quantization
- FP8 + MXFP4
- Reasoning
- off
- Profile
- quick
- Suite
- v2b
- Date
- 2026-09-26 – 2026-09-27
- Hardware
- 2× NVIDIA DGX Spark, je 128 GB RAM
- Inference software
- vLLM 0.26.0
- Measured
- directly at the inference server
- Overall
- 59.4
Categories
| Category | Score |
|---|
| Knowledge & reasoning | 69.7 |
| Math | 41.6 |
| Instruction following | 49.8 |
| Code | 61.5 |
| German | 74.5 |
Tasks
| Category | Task | Score | ± | n |
|---|
| Knowledge & reasoning |
mmlu_pro |
82.0 |
± 2.2 |
300 |
| Knowledge & reasoning |
gpqa_diamond_gen |
68.7 |
± 3.3 |
198 |
| Knowledge & reasoning |
musr_gen |
58.4 |
± 3.1 |
250 |
| Math |
omni_math_gen |
41.6 |
± 3.1 |
250 |
| Instruction following |
ifbench |
37.0 |
± 2.8 |
300 |
| Instruction following |
ifeval |
88.7 |
± 2.6 |
150 |
| Code |
livecodebench_gen |
61.5 |
± 3.4 |
200 |
| German |
mmlu_prox_de |
78.7 |
± 2.4 |
300 |
| German |
include_de_gen |
66.2 |
± 3.8 |
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] |
– |
– |
27.1 |
25.4 |
28.2 |
5 |
| Prompt processing [t/s] |
512 |
525 |
959.7 |
957.4 |
966.7 |
5 |
| Prompt processing [t/s] |
4096 |
4024 |
2177.6 |
2132.8 |
2178.6 |
5 |
| Prompt processing [t/s] |
16384 |
16036 |
2097.1 |
2093.4 |
2106.5 |
5 |
| Time to first token [ms] |
512 |
525 |
546.0 |
543.1 |
548.4 |
5 |
| Time to first token [ms] |
4096 |
4024 |
1847.9 |
1847.0 |
1887.0 |
5 |
| Time to first token [ms] |
16384 |
16036 |
7646.6 |
7612.6 |
7660.1 |
5 |
Measured directly at the inference server.