SenseNova-U1.5

SenseTime's open unified visual model, successor to NEO: an 8B mixture-of-transformers model that understands, reasons about, and generates images inside one encoder-free, VAE-free architecture, with a spatially coherent patch-reconstruction visual interface and native resolutions up to 4K. Post-training trains separate experts for aesthetics, bilingual text rendering, infographics, and editing, then consolidates them through multi-expert on-policy distillation. Released under Apache 2.0 with Preview, SFT, and LoRA siblings and a promise of open SFT, RL, and distillation code; the 65-author report is dated 10 September 2026 and the GitHub repository has passed 6,600 stars.