Compact and efficient, this model punches at its weight class for basic instruction-following tasks like summarization, classification, and simple Q&A. At 1.2B parameters, it trades reasoning depth and nuanced understanding for speed and low resource consumption — expect occasional gaps in complex logic or multi-step tasks. It runs comfortably on modest hardware, making it a practical choice for edge deployments or high-throughput pipelines.
| Benchmark | Score | Type | Recorded |
|---|---|---|---|
| GPQA Diamond | 3.4 | accuracy | 26d ago |
| MATH | 7.0 | accuracy | 26d ago |
| IFEval | 57.0 | accuracy | 26d ago |
| BBH | 8.7 | accuracy | 26d ago |
| MMLU-Pro | 7.6 | accuracy | 26d ago |
| MuSR | 3.0 | accuracy | 26d ago |