SenseTime

Native unified multimodal model open-sourced August 20, 2026 on the NEO-unify architecture (Apache 2.0): native 4K generation, stronger complex-instruction following, text/layout rendering and image editing, with support for 3-4K-character ultra-long instructions.

Flagship delivery-grade natively multimodal agent foundation unveiled July 18, 2026 at WAIC, built for long-horizon tasks such as data analysis, deep research and iterative image editing (API/invite at launch).

Unified vision understanding-and-generation model fully open-sourced July 13, 2026: the 7B Mixture-of-Transformers model reformulates detection, segmentation and depth estimation as multimodal text generation, shipped with a 50M-pair training corpus.

Sensenova V6 Turbo is an full modal model from Sensetime, designed for general AI application workflows.

Sensenova V6 Pro is an full modal model from Sensetime, designed for general AI application workflows.

Sensenova V6 Reasoner is an reasoning model model from Sensetime, designed for reasoning, planning, and complex problem solving.

SenseNova V6.5 Miaohua Base is an image generation model from Sensetime, designed for image generation and creative visual workflows.

SenseNova V6.5 Miaohua Pro is an image generation model from Sensetime, designed for image generation and creative visual workflows.

Natively unified understanding-and-generation multimodal series released open-source in late April 2026 alongside a free Token Plan; iterated as the open U1.5-Lite-Preview toward U1 Pro.