Kapitsa Super
Nemotron 3.5 Super VL 120B-A12B (NVFP4, 76.4 GB) running locally on Apple Silicon through Kapitsa. A vision model for 96 GB+ Macs.
kapitsa serve --model super
Speed
| Mac | Decode (tok/s) |
|---|---|
| M5 Max 128 GB | 87 |
Architecture
Hybrid Mamba-2 MoE, 120B total parameters, 12B active per token, NVFP4 quantization. Supports text and image inputs.