We design CRM, CX, EX, and AI strategies that align systems, people, and data for measurable results. This is where we publish our open model work.
Near-lossless 8-bit build at 8.5 GB. Runs on a single consumer GPU with the widest runtime support of the three.
4-bit build in NVIDIA's Blackwell-native format, with the full module inventory and calibration manifest published alongside the weights.
Group-quantized build for Macs, chosen from an on-device recipe sweep published in the model card.
Each build ships with reproducible calibration data, eval pins, and raw results. Validated on RTEB-open within 0.008 nDCG@10 of NVIDIA's published scores. Browse the full collection.