A compact, no-frills text model that handles straightforward language tasks with reasonable fluency. At 6.7B parameters it works within tight compute budgets, though it shows its limits on complex reasoning or nuanced instruction-following compared to larger siblings. It's the base pretrained version — meaning it responds more like a text-completer than a chat assistant out of the box.
| Benchmark | Score | Type | Recorded |
|---|---|---|---|
| IFEval | 25.2 | accuracy | 2d ago |