SenseTime co-founder and chief scientist Lin Dahua admitted in a CNBC interview that the company's latest model, SenseNova U1, still lags behind OpenAI's GPT Image 2 and Google's Gemini Nano Banana in image generation capabilities — but its generation cost is only one-tenth that of ChatGPT Images 2.0. Lin's strategy boils down to: "Many scenarios don't need the top-tier model; adequate is enough."
SenseNova U1 is a native multimodal model that unifies understanding and generation, built on SenseTime's proprietary NEO-unify architecture. By integrating language and vision processing into a single system, it eliminates conversion steps between modalities, improving speed and efficiency. SenseTime drew inspiration from DeepSeek's approach to building high-performance models under constraints of limited funding and compute, making low cost its core competitive advantage. Lin revealed that ByteDance's video model Seedance once posed competitive pressure, but SenseTime has since integrated some of Seedance's capabilities into its short-video tool Seko.
On the financial front, SenseTime narrowed its net loss by 58.6% in 2025, and EBITDA turned positive for the first time in the second half of the year — a first since the company's IPO in 2021. In an April 28 research note, Jefferies highlighted four challenges facing pure-play AI model companies: low customer loyalty, limited differentiation, a crowded field, and high training costs. Vey-Sern Ling, senior equities advisor at UBP, noted that platform companies like Alibaba, Tencent, and ByteDance can subsidize AI development with cash flow from their core businesses, while independent AI firms continue to bleed money. Facing US sanctions, SenseTime has shifted its overseas focus to Southeast Asia, Northeast Asia, the Middle East, and Brazil.