Apriel
modelServiceNow's enterprise SLM family. Apriel-5B: 4.8B parameters trained on 4.5T tokens. Outperforms OLMo-2-7B and Mistral-Nemo-12B despite being smaller.
Apriel-Nemotron-15B-Thinker (August 2025, with NVIDIA): 15B reasoning model that matches o1-mini and QwQ-32B at half the memory. Apache 2.0.
Model Details
Architecture DENSE
Parameters 4.8B
Variants
| Name | Parameters | Notes |
|---|---|---|
| Apriel-5B | 4.8B | — |
| Apriel-Nemotron-15B-Thinker | 15B | Reasoning model with NVIDIA |