Guanlan AI ups accuracy, cuts cost

Hikvision said on September 7 that its Guanlan AI application stack now delivers higher accuracy, lower deployment cost and more natural interaction in video systems used for surveillance and other AIoT tasks. The statement came in a PR Newswire release from Hangzhou, China, more than a year after Hikvision introduced Guanlan large-scale AI models in April 2025. Hikvision said the newer applications can interpret live scenes, search recorded footage and respond to user prompts in natural language. (prnewswire.com) The company described that shift as a move from video systems that merely detect events to systems that can work with operators in a more conversational way. (hikvision.com) ### What, exactly, did Hikvision say has improved? The September 7 release said Guanlan AI applications now combine “higher accuracy, lower cost, smarter interaction” in commercial video deployments. Hikvision said the gains come from its large-scale models and from application-level tuning that lets systems match specific operating scenarios with less manual work. Hikvision’s official Guanlan technology page says the model suite is built around computer vision, natural language processing and multimodal fusion. The company says those models are designed for AI-powered Internet of Things uses, including video understanding, search and automated response across edge devices and broader systems. (prnewswire.com) ### How does “smarter interaction” work in practice? Hikvision said the practical change is that users can query recorded video in ordinary language instead of relying only on fixed filters, timestamps or manual review. In the September 7 release, the company said today’s systems can identify what they are looking at, recognize what has already been recorded and interpret what the user wants through natural-language interaction. (prnewswire.com) A Hikvision demonstration for AcuSeek, one of the company’s search tools, shows that approach in product form. Hikvision said Open Search powered by Guanlan lets users retrieve video footage by typing a sentence, rather than building a query through conventional surveillance software menus. (hikvision.com) ### Where does Guanlan fit in Hikvision’s product strategy? Hikvision launched Guanlan large-scale AI models on April 29, 2025, describing them as pre-trained on industry knowledge and built for AIoT products and applications. The company said at launch that the models were intended to support products at the edge while also improving broader system intelligence. (prnewswire.com) Hikvision’s current technology materials present Guanlan as a platform layer rather than a single feature. The company says the models support applications spanning security protection, business intelligence, traffic management and other operational settings where video analysis and system response are tied together. (youtube.com) ### What does Hikvision mean by lower cost? The September 7 release did not disclose pricing, benchmark data or named customer deployments. Hikvision instead said lower cost comes from reducing manual work, improving fit between model behavior and deployment scenario, and making systems safer and more efficient to operate. (prnewswire.com) Because Hikvision did not publish comparative metrics in the release, the company’s cost claim is best read as a product statement rather than a quantified earnings or procurement figure. The official materials available on September 8 focus on capabilities, model design and use cases, not audited savings. (hikvision.com) ### What should readers watch next? Hikvision’s next public updates are likely to appear first on its official Guanlan technology pages and product demonstrations. As of September 8, those pages describe the model suite, cite its April 2025 launch and show how tools such as AcuSeek use natural-language search on recorded video. (prnewswire.com) (hikvision.com)