MOSI, a Shanghai-based AI startup, has raised over RMB 1 billion ($150 million) in a Pre-A funding round. The company is focusing on real-time omnimodal interaction, a new frontier in AI beyond traditional text-based models.
TL;DR
- MOSI raises $150M to develop real-time omnimodal AI models, challenging China's established foundation model players.
- The funding will support R&D in real-time interaction and global commercialization, with a focus on speech, vision, and multimodal capabilities.
- MOSI's approach differs from competitors like DeepSeek and Moonshot AI, which focus primarily on text-based reasoning and coding.
What happened
MOSI, formally known as Shanghai Mosi Intelligence Technology, announced a Pre-A funding round exceeding RMB 1 billion ($150 million) on October 12. The round was led by Huakong Fund, with significant participation from CDB Technology Innovation, Shanghai Science and Technology Innovation Fund, and other major investors.
The company, spun off from the OpenMOSS research team at Fudan University, has shifted its focus from text-based models to speech, vision, and real-time interaction. MOSI has already released over 20 open-source models in 2024, including MOSS-TTS for speech generation and MOSS-VL for visual understanding.
MOSI's upcoming MOSS Live system aims to integrate these models, enabling continuous real-time interaction. The funding will also support MOSI's 'AI for AI' initiative, which uses generative models to produce and evaluate training data, reducing manual labor.
Why it matters
MOSI's focus on real-time omnimodal interaction sets it apart from competitors like DeepSeek and Moonshot AI, which concentrate on text-based reasoning and coding. This differentiation could attract developers and startups looking for advanced multimodal capabilities.
The substantial Pre-A funding round indicates strong investor confidence in MOSI's unique approach and the potential of real-time omnimodal AI. This could encourage further innovation and competition in the Chinese AI landscape.
For investors, MOSI's focus on a niche but growing segment of the AI market presents an opportunity to support a company that could challenge the dominance of established players. However, the competitive landscape and MOSI's ability to execute on its vision remain key considerations.
Key facts
- MOSI raised over RMB 1 billion ($150 million) in a Pre-A funding round led by Huakong Fund.
- The company was spun off from the OpenMOSS research team at Fudan University, led by professor Qiu Xipeng.
- MOSI has released over 20 open-source models in 2024, focusing on speech, vision, and real-time interaction.
- The upcoming MOSS Live system aims to integrate MOSI's models for continuous real-time interaction.
- Major investors include CDB Technology Innovation, Shanghai Science and Technology Innovation Fund, and New Alliance Capital.
- MOSI's 'AI for AI' initiative uses generative models to produce and evaluate training data.
- The funding will support R&D in real-time omnimodal interaction and global commercialization.
- MOSI's approach differs from competitors like DeepSeek and Moonshot AI, which focus on text-based reasoning and coding.
Context
China's foundation model landscape has seen significant investment, with companies like DeepSeek and Moonshot AI attracting substantial funding. However, MOSI's focus on real-time omnimodal interaction represents a new frontier in AI development.
The growing demand for AI capabilities beyond traditional text-based models is driving innovation in speech, vision, and multimodal interaction. MOSI's approach could appeal to developers and startups looking for advanced capabilities in these areas.
The competitive landscape in Chinese AI remains dynamic, with new entrants like MOSI challenging established players. The success of MOSI's unique approach could influence the direction of AI development in China and beyond.
