ZKH Group and Intel have introduced LingLong Insight 1.0, a 4-billion-parameter multimodal AI model designed to verify industrial supplies directly at factory and warehouse workstations. Developed by ZKH’s AI subsidiary, LingLong Intelligence, the model combines image recognition, text understanding and industrial product knowledge to identify materials and detect specification mismatches against delivery documents. Running locally on Intel Core Ultra processors through the LingLong Agent Box, the system is designed to reduce cloud dependence and support faster on-site decisions. The launch highlights the growing role of edge AI in manufacturing, industrial automation and supply chain operations.
ZKH and Intel Bring Multimodal AI to Industrial Worksites
ZKH Group is expanding its industrial artificial intelligence capabilities through a collaboration with Intel, introducing an edge-based multimodal AI model designed to address a practical problem in manufacturing and maintenance operations: verifying whether delivered components match their documented specifications.
Announced at Intel Connection 2026 in Suzhou, China, LingLong Insight 1.0 combines image and text understanding with industrial product knowledge. The model is developed by LingLong Intelligence, ZKH’s AI subsidiary, and uses Intel’s edge computing technology to perform verification locally rather than depending entirely on cloud-based processing.
The initial application focuses on goods receipt, where employees must identify industrial supplies, inspect product labels and compare specifications against purchase and delivery documentation.
Industrial Verification Becomes an AI Application
Industrial procurement involves thousands of components that can look nearly identical but differ in dimensions, materials, performance ratings or compatibility requirements.
For example, two bearings may share a similar appearance while having different specifications that determine whether they can be installed in a particular machine. An incorrect identification can introduce inventory errors, delay maintenance or interrupt production.
LingLong Insight 1.0 is designed to help address this challenge through a visual verification workflow.
Industrial cameras capture images of delivery documents, packaging and product labels. The model then identifies the supplied items, interprets relevant text and checks product specifications against the accompanying documentation. When it detects inconsistencies, it flags the discrepancies and displays the associated images for worker review.
This combination of visual recognition and document understanding makes the application different from conventional industrial image-classification systems, which may identify an object without establishing whether it meets a particular operational specification.
The model is intended to connect visual information with structured industrial knowledge, allowing it to evaluate products in the context of their intended use.
A 4-Billion-Parameter Model Designed for Edge Deployment
LingLong Insight 1.0 contains 4 billion parameters and is designed to operate on the LingLong Agent Box, which uses Intel Core Ultra processors combining CPU, GPU and neural processing unit (NPU) capabilities.
The deployment architecture places AI inference close to the industrial workstation where images and documents are generated.
This matters because industrial environments often require responsive decisions without depending on continuous communication with remote cloud infrastructure. Local inference can reduce network latency, limit the movement of sensitive operational information and support applications in environments where connectivity is inconsistent.
Intel’s heterogeneous computing architecture is relevant here because different processing units can handle different workloads. CPUs manage general computing tasks, GPUs accelerate parallel processing, and NPUs are designed to execute AI workloads efficiently.
However, actual performance depends on model optimization, processor configuration, memory availability and workload characteristics. ZKH has not disclosed independent inference benchmarks, processing latency figures or detailed hardware configurations for LingLong Insight 1.0.
The model’s 4-billion-parameter size also places it within the broader movement toward smaller, specialized AI models that can run closer to data sources instead of relying exclusively on large cloud-hosted foundation models.
Industrial Data Provides the Domain-Specific Context
Beyond computing hardware, ZKH is positioning its industrial data as a key part of the model’s development.
The company says it has accumulated nearly 30 years of industrial supplies experience, more than 27 million product records and approximately 4 billion product parameters. LingLong Intelligence has standardized this information and combined it with industry expertise to create structured data for industrial AI applications.
This is important because general-purpose multimodal models may recognize an industrial component without understanding the precise specifications that determine its suitability for a particular machine or application.
Domain-specific product records can provide the additional context required to distinguish between similar items. In this case, the model’s value depends not simply on visual recognition but on connecting what it sees with relevant product specifications and procurement documentation.
The approach also creates opportunities beyond goods-receipt inspection. ZKH plans to extend LingLong Insight to warehouse inbound and outbound verification, production-line material checks, and equipment operation and maintenance.
If these applications prove effective, the technology could become part of a wider industrial AI workflow spanning procurement, inventory management and production operations.
Edge AI Adoption Is Expanding Across Manufacturing
The launch comes as manufacturers increasingly explore AI applications that interact directly with physical operations.
Cisco’s 2026 State of Industrial AI research found that 61% of surveyed organizations were using AI in live industrial operations, while 20% reported scaled, mature deployments. The research also found that 44% of manufacturing respondents expected greater edge computing capacity to be necessary as AI workloads expand. Cisco Newsroom+1
These findings highlight why edge computing is becoming an important part of industrial AI infrastructure. Visual inspection, equipment monitoring, material verification and safety applications often require decisions close to machines, production lines and warehouse workstations.
The market is also evolving beyond traditional machine vision. Multimodal AI can combine images with accompanying text, technical specifications and operational records, allowing industrial systems to evaluate more complex situations.
This development is relevant to companies including NVIDIA, Microsoft and Intel, which are investing in different layers of industrial AI infrastructure, from accelerated computing and AI software to edge processors and enterprise deployment platforms.
Intel’s involvement gives LingLong Intelligence access to its processor ecosystem, while ZKH contributes specialized industrial data and procurement expertise. Their collaboration reflects a broader industry direction in which computing hardware and domain-specific AI knowledge are developed together.
From Goods Receipt to Industrial Automation
The immediate commercial application is relatively focused: helping workers verify incoming industrial supplies before confirming receipt. Its longer-term potential lies in extending the same intelligence to additional operational processes.
Warehouse inspections could use similar capabilities to compare outgoing materials with order records. Production-line applications could verify whether the correct components are being supplied to manufacturing equipment. Maintenance workflows could eventually use visual and textual information to support equipment inspections and component identification.
ZKH and Intel also intend to deepen their collaboration and explore international market opportunities. However, the companies have not announced specific overseas deployments, customer commitments or a commercialization timetable for these planned applications.
The competitive challenge will involve demonstrating consistent recognition accuracy across different industrial environments, product categories, lighting conditions and documentation formats. Integration with existing warehouse management, ERP and manufacturing execution systems will also be important.
LingLong Insight 1.0 illustrates how industrial AI is moving toward specialized, locally deployed multimodal systems. Rather than treating factories as environments that simply consume general-purpose AI services, the approach brings models closer to operational data and gives them industry-specific knowledge.
For ZKH, the opportunity is to extend its MRO procurement expertise into AI-powered industrial workflows. For Intel, the collaboration provides another application of edge computing in manufacturing. The technology’s broader significance will depend on whether it can turn product verification into a reliable foundation for more extensive industrial automation.
Market Landscape
Industrial AI is evolving from conventional automation and machine vision toward multimodal systems capable of interpreting images, documents and operational information together.
Three developments are shaping this market:
- Edge AI infrastructure: Manufacturers are moving inference closer to production environments to reduce latency, cloud dependency and data-transfer requirements.
- Industrial multimodal AI: Vision-language models are expanding AI applications from object detection to specification verification, document interpretation and quality inspection.
- Domain-specific AI models: Specialized industrial datasets and product knowledge are becoming important differentiators for AI systems operating in complex manufacturing environments.
Cisco’s 2026 industrial research found that 61% of surveyed organizations were already using AI in live industrial operations. Meanwhile, Gartner’s August 2026 research identified industrial AI as a potential $300 billion manufacturing technology opportunity. Cisco Newsroom+1
ZKH and Intel are positioning LingLong Insight within this expanding market through a combination of specialized industrial knowledge and edge computing.
The competitive landscape includes NVIDIA’s industrial AI ecosystem, Intel’s edge computing platforms, and Microsoft Azure’s industrial AI capabilities. The differentiation will increasingly depend on deployment efficiency, domain-specific accuracy, hardware compatibility and integration with existing operational systems.
Top Insights
- ZKH and Intel launched LingLong Insight 1.0, a 4-billion-parameter multimodal AI model designed for industrial verification and edge deployment.
- The model combines image recognition, text understanding and industrial product knowledge to identify specification mismatches during goods receipt.
- LingLong Insight runs locally on LingLong Agent Box using Intel Core Ultra processors, reducing reliance on cloud-based inference.
- ZKH contributes nearly 30 years of industrial expertise, more than 27 million product records and approximately 4 billion product parameters.
- Planned applications include warehouse verification, production-line material checks and equipment operation and maintenance.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
