Neutrino-1 8B: A Single Model for Datacenter GPUs and Laptops

We introduce Neutrino-1 8B, a transformer model using a proprietary ternary format to shrink its footprint to just 3.88 GB. This design allows the same artifact to run efficiently on everything from H100 datacenter GPUs to 16 GB MacBook laptops without conversion. By decoding weights directly inside matrix kernels, we achieve high throughput and enable speculative decoding on consumer hardware while maintaining bit-exact performance across all platforms.