The race for edge AI is shifting from multi-billion parameter models to compact, specialized transformers that can run locally on low-end laptops and mobile phones.
Specification Comparison
| Model | Parameters | FP16 Disk Size | Math Execution Accuracy |
|---|---|---|---|
| SmolLM-135M | 135 Million | ~270 MB | 62.5% (Parametric) |
| Nova-Mini V1 | 57.2 Million | ~109 MB | 100% (PoT Tool-Calling) |
The Tool-Calling Advantage
Rather than relying on noisy parametric weights for arithmetic, Nova-Mini delegates math directly to an internal Python REPL, achieving 100% exact match on arithmetic benchmark sets.