SenseNova-U1: Unifying Multimodal Understanding and Generation with NEO-Unify Architecture

SenseNova released SenseNova-U1 with 4 MoT multimodal models. The post lists 8B and A3B variants with GitHub and HuggingFace weight links. The key claim is a monolithic architecture instead of adapters; benchmarks are not disclosed.

Why it matters: HKR-H/K/R pass: open weights, 4 MoT multimodal models, and a single architecture are concrete. Benchmarks are not disclosed, and the source is Reddit, so this stays in the 72–77 featured-threshold band.