A compact, instruction-tuned model that punches at its weight class for simple conversational tasks and basic text generation. Its 2.5B parameter footprint means it runs comfortably on modest hardware, but it shows its limits quickly when reasoning chains get long or context grows complex. Think of it as a reliable lightweight — fast and accessible, but not the one you'd hand a multi-step problem.
| Benchmark | Score | Type | Recorded |
|---|---|---|---|
| IFEval | 30.7 | accuracy | 6d ago |