Chinese AI labs are increasingly releasing very serious open source work.
Rohan Paul Twitter · Rohan Paul (@rohanpaul_ai) · 2026-05-20
Chinese AI lab SenseTime releases SenseNova U1 on HuggingFace, a native multimodal model using a Mixture-of-Transformers architecture with 38 billion total parameters and 3 billion active, targeting readable and consistent image-text generation.
Appears in
Extraction
Topics: open-source-aichinese-ai-labsmultimodal-modelsimage-generation
Claims
- Chinese AI labs are releasing open-source models competitive in quality with leading international labs.
- SenseNova U1 uses a Mixture-of-Transformers (MoT) architecture with 38B total parameters and 3B active parameters via MoE.
- SenseNova U1 specifically targets readable, structured, and consistent image-text generation, which Rohan Paul identifies as the hardest sub-problem in image generation.
Key quotes
Chinese AI labs are increasingly releasing very serious open source work.
It attacks the hardest part of image generation: readable, structured, consistent image-text