--- title: "SenseTime, running at the 'super speed' of China's AI." type: "Topics" locale: "en" url: "https://longbridge.com/en/topics/22286605.md" description: "SenseTime is demonstrating the true meaning of 'If you can be new for one day, you can be new every day, and then be new again' with its own actions. The release of China's first streaming multimodal interaction large model 'Rixin 5.5' is clear evidence. It's hard to imagine that in just two months, SenseTime has achieved such significant evolution. Remember, just two months ago, SenseTime released 'Rixin 5.0', which was China's first domestic large model comparable to GPT-4 Turbo. Now, SenseTime is coming strong with 'Rixin 5.5', once again showing us the 'super speed' of China's original AI..." datetime: "2024-07-06T06:06:39.000Z" locales: - [en](https://longbridge.com/en/topics/22286605.md) - [zh-CN](https://longbridge.com/zh-CN/topics/22286605.md) - [zh-HK](https://longbridge.com/zh-HK/topics/22286605.md) author: "[产业深观](https://longbridge.com/en/profiles/11766346.md)" --- > Supported Languages: [简体中文](https://longbridge.com/zh-CN/topics/22286605.md) | [繁體中文](https://longbridge.com/zh-HK/topics/22286605.md) # SenseTime, running at the 'super speed' of China's AI. SenseTime is demonstrating the true meaning of "If you can renovate yourself one day, do so from day to day" with its actions. The release of China's first streaming multimodal interactive large model, "SenseNova 5.5," is clear evidence of this. It's hard to imagine that in just two months, SenseTime has achieved such significant evolution. Just two months ago, SenseTime's "SenseNova 5.0" was China's first domestic large model benchmarked against GPT-4 Turbo. Now, SenseTime has arrived with "SenseNova 5.5," once again showcasing the "hyper-speed" of China's original AI. _**The performance of the 600-billion-parameter base model has been comprehensively upgraded. Extensive use of synthetic high-order chain-of-thought data enhances reasoning capabilities, with significant improvements in mathematical logic, English, and instruction-following abilities.**_ _**Pioneering China's first "what-you-see-is-what-you-get" model, "SenseNova 5o," featuring streaming multimodal interaction, introduces a new AI interaction paradigm.**_ _**The edge-side model has been fully upgraded with the release of "SenseNova 5.5 Lite," achieving a 10% improvement in model accuracy, a 15% boost in inference efficiency, and a 40% reduction in first-packet latency compared to the April 5.0 version.**_ These are the key updates in "SenseNova 5.5." Through these advancements, we not only witness SenseTime's hyper-speed evolution but also see it leading China's original AI to new heights. If we were to define SenseTime's "SenseNova 5.5," it is China's first large model with native streaming multimodal interaction capabilities. _**From being China's first domestic large model benchmarked against GPT-4 Turbo two months ago to now achieving parity with GPT-4o in interaction effects and multiple core metrics, SenseTime once again demonstrates the "hyper-speed" of AI through action.**_ **Honoring Xiaoou Tang: SenseTime Reveals the Foundation of "Hyper-Speed"** Every evolution has an origin; every phenomenon has an explanation. If we must identify the underlying reason for SenseTime's AI "hyper-speed," it lies in its long-term commitment to AI, patience, years of technological accumulation, talent strategy, and dedication to originality. This is why, before releasing "SenseNova 5.5," SenseTime's Chairman and CEO Xu Li used the company's controllable video-generation model Vimi to create a tribute video honoring SenseTime's founder, Professor Xiaoou Tang. As Xu Li said, "Without SenseTime's long-term technological accumulation and talent strategy, there would be no today's achievements." Perhaps this is the true foundation of SenseTime's AI "hyper-speed." It is the relentless pursuit of long-termism and forward-looking strategies by pioneers like Professor Tang that has enabled SenseTime to repeatedly become the "first" and showcase the momentum of China's original AI. If SenseTime's leaders have charted its course, the company's strategic layout has provided the "hard power" for continuous iteration. From data to algorithms, computing power, and from products to industries and scenarios, SenseTime is turning AI ideals into reality step by step. Massive data, powerful algorithms and computing, a high-quality product ecosystem, and rich industry applications define SenseTime today. The latest evolution of SenseNova further reveals the formidable strength derived from these internal forces. Viewing SenseTime's progress from this perspective, the reasons behind its "hyper-speed" become clearer and more natural. Examining "SenseNova 5.5's" multimodal capabilities from this angle explains its parity or even superiority over GPT-4o in core benchmark metrics. "SenseNova 5.5" shows marked improvements in mathematical reasoning, English proficiency, and instruction-following, achieving parity with GPT-4o in interaction effects and multiple core metrics. ![Image](https://pub.pbkrs.com/social/2024/11766346/uK5mACPtymqoMhyGqpDkJ7zKDQC8VWiw.jpg?x-oss-process=style/lg) ![Image](https://pub.pbkrs.com/social/2024/11766346/ktp4V596Cp9iC3YcWAB4tkscVqFYtvWW.jpg?x-oss-process=style/lg) Additionally, "SenseNova 5.5" adopts a hybrid cloud-edge-expert architecture to maximize synergy, reduce inference costs, and is trained on over 10TB of high-quality tokens, including extensive synthetic chain-of-thought data to enhance reasoning. Clearly, "SenseNova 5.5" is becoming the new technological foundation. This foundation enables SenseTime's outstanding performance in "SenseNova 5o" and "SenseNova Lite." **Natural and Smooth: SenseTime Delivers "What You See Is What You Get" AI** As the AI large model industry enters deeper waters, innovative interaction modes that make AI as easy to use as the internet have become key to leading the industry. SenseTime's leadership in innovative interaction modes is another testament to its AI "hyper-speed." _**By integrating cross-modal information—voice, text, images, and video—"SenseNova 5o" delivers real-time streaming multimodal AI interaction.**_ The experience is as direct as human communication, enabling instant understanding of what you see and need. This interaction mode excels in multi-task adaptability, naturally handling diverse tasks within a single model and adjusting behavior and output based on context. _**From scene analysis and object description to book summaries, rough sketches, and facial emotions, "SenseNova 5o" handles them all with precision, smooth interaction, and even playful banter.**_ SenseTime's "SenseNova 5o" offers a seamless and natural interaction experience, making AI feel like conversing with a real person. Thanks to SenseTime's leadership in innovative interaction, using AI like the internet has become a reality. Few players in today's AI large model market can match SenseTime's natural and smooth experience. Live demonstrations vividly illustrate this. From playful greetings to accurately identifying locations and summarizing random book pages, "SenseNova 5o" provides deep understanding, not mere recognition. _**SenseTime's "SenseNova 5o" delivers not just real-time speed but also the natural fluidity of daily life.**_ With such smoothness, using AI is no longer a high-barrier task but a natural and commonplace activity. Only with such interaction can AI become as effortless as scanning a QR code to pay. As the AI large model market intensifies and players pursue commercial perfection, SenseTime's "SenseNova 5o" leads with outstanding innovation, challenging GPT-4o once again. **Edge-Side Model Upgrade: SenseTime Drives Down Costs** For AI large models to reach the masses, they must leverage one terminal after another. Currently, SenseTime's edge-side models serve over 150+ commercial clients across smartphones, tablets, VR headsets, car systems, smart lamps, and IoT devices. Even a smart lamp can now connect with SenseTime's AI large models, thanks to the upgraded "SenseNova 5.5 Lite." It offers better performance, faster inference, and a more robust edge-side model matrix. "SenseChat Lite-5.5" is the most comprehensive edge-side model to date, complemented by cloud-edge synergy for balanced performance and speed. ![Image](https://pub.pbkrs.com/social/2024/11766346/p3KEzjvAvPa9Wa4m8PegW424n9fw7jw9.jpg?x-oss-process=style/lg) But the key lies elsewhere. SenseTime not only upgraded its edge-side models but also slashed prices to just 9.9 RMB per device per year, making AI accessible to all. With continuous edge-side upgrades, SenseTime's product ecosystem has expanded. From SenseChat to Xiao Huan Xiong, from human-like dialogue models to finance, agriculture, tourism, and healthcare, SenseTime has built a vast application landscape. Without edge-side advancements, such widespread adoption would be impossible. _**SenseTime's technological progress is dismantling AI's price barriers, enabling broader access to AI's benefits.**_ Thus, SenseTime achieves not just technological "hyper-speed" but also industry and scenario adoption "hyper-speed." **Conclusion** "If you can renovate yourself one day, do so from day to day"—SenseTime embodies this perfectly. Its relentless innovation has generated the energy for "hyper-speed," leading to rapid evolution in just two months. From China's first GPT-4 Turbo benchmark to now challenging GPT-4o in streaming multimodal interaction, SenseTime represents China's competitive stance in global AI. SenseTime's small steps drive China's AI leaps forward. Thanks to pioneers like SenseTime, China maintains its place and voice in global AI.$SENSETIME-W(00020.HK) **—End—** ### Related Stocks - [SENSETIME-W (00020.HK)](https://longbridge.com/en/quote/00020.HK.md) - [SENSETIME-WR (80020.HK)](https://longbridge.com/en/quote/80020.HK.md) ## Comments (1) - **拂袖斋客人 · 2024-07-08T08:47:36.000Z**: Afraid