MOSI
MOSI Intelligence builds contextual, multimodal AI that connects AI, humans, and the physical world. Its work spans foundation models and products for speech, audio, vision, content generation, and real-time interaction beyond the chat window.
At a Glance
- Developers and enterprises integrating AI APIs
- Content creators and media/entertainment teams
- Customer-service and cross-language communication operations
- Automotive and smart-cockpit manufacturers
- +2 more
AI Tools by MOSI
(1)
Moss API Platform
Image and Speech Generation API
Discussions
No discussions yet
Be the first to start a discussion about MOSI
Latest News
MOSS AI raises over RMB 1 billion in Pre-A funding for next-generation foundation models and real-time omnimodal interaction.
Qiu Xipeng attends Huawei HC Conference and is appointed an Ascend MVP; MOSI highlights Huawei/Ascend collaboration and 29 ms first-packet speech latency.
MOSS appears at the 2026 Inclusion · Bund Summit, showcasing real-time model interaction experiences and MOSS Create, MOSS API, and MosMos.
Co-founder and CEO Li Shimin speaks at ChinaJoy on connecting AI, humans, and the physical world.
Products & Services
Usage-based API platform exposing MOSS capabilities including speech generation, long-form audio transcription, audio understanding, video understanding, voice design, and multimodal model access through API keys and documented REST endpoints.
One-stop AI creation platform for creators, offering voice cloning, timbre/voice design, video dubbing, and AI podcast creation.
Full-duplex real-time interaction system that listens while speaking, handles semantic turn-taking and interruptions, coordinates foreground conversation with background reasoning and tool calls, translates while listening, sees video or shared screens, and takes actions.
Desktop voice-input product that lets users press Fn to speak in any application and place the transcribed text at the cursor.
Market Position
MOSI positions itself around contextual, real-time omnimodal interaction rather than only text chat, coding, or static content generation. Its differentiation is the combination of open multimodal foundation models, continuous speech/video understanding, full-duplex interaction, and products/APIs intended to connect AI with physical-world devices and situations. Relevant comparison points include other speech and multimodal model/API providers such as OpenAI, Google, Alibaba/Qwen, MiniMax, Fish Audio, CosyVoice, and ElevenLabs; MOSI emphasizes open-source accessibility and real-time context handling.
Leadership
Founders
Qiu Xipeng
Founder and chief scientist of MOSI Intelligence; distinguished professor at Fudan University and vice dean of the Shanghai Innovation Institute. He began researching pretraining in 2016, led the OpenMOSS team, and led the release of MOSS, described by MOSI as China's first open-source ChatGPT-like model, in February 2023.
Li Shimin
Co-founder and CEO of MOSI Intelligence. He is a core author of the SpeechGPT paper and has represented the company publicly on contextual intelligence and next-generation human-computer interaction.
Executive Team
Li Shimin
Co-founder and CEO
Core author of the SpeechGPT paper; public spokesperson for MOSI's contextual-intelligence and real-world interaction strategy.
Qiu Xipeng
Founder and Chief Scientist
Fudan University distinguished professor and Shanghai Innovation Institute vice dean; leader of the OpenMOSS team and original MOSS release.
Founding Story
MOSI Intelligence emerged in late 2024 from the OpenMOSS research team led by Fudan professor Qiu Xipeng, with support from the Shanghai Innovation Institute's talent, computing resources, and academic-industry environment. The initial research trajectory moved from the MOSS language model to SpeechGPT and AnyGPT, then into focused engineering of speech, multimodal understanding, generation, and real-time interaction systems.
Business Model
Revenue Model
Usage-based API billing by actual consumption, with enterprise solutions offering custom quotas, dedicated service, SLA, and technical support; consumer/creator products also provide online experiences and creation workflows.
Pricing Tiers
40 credits per 10K input characters; each request is billed by actual usage.
16 credits per hour of input audio/video.
40 credits per hour of input audio/video.
Custom quotas, dedicated service, SLA, and technical support.
Target Markets
- Developers and enterprises integrating AI APIs
- Content creators and media/entertainment teams
- Customer-service and cross-language communication operations
- Automotive and smart-cockpit manufacturers
- Wearable, toy, home-device, and robotics companies
- Research and open-source AI communities
- Voice assistants and conversational AI
- In-car and smart-cockpit interaction
- AI glasses, toys, companion devices, and robots
- Meetings, interviews, podcasts, and multi-speaker transcription
- Livestream, sports, monitoring, and continuous video understanding
- Audiobooks, narration, games, film/TV dubbing, and short-form video
- MOSS Create attracted more than 100,000 active creators during closed beta