BEIJING, Oct. 6, 2025 /PRNewswire/ -- Recently, MiningLamp Technology, a leading enterprise in China's enterprise-level large models and data intelligence sector, officially launched its specialized large model product line DeepMiner. Targeting business data analysis, DeepMiner aims to provide enterprises with more efficient, verifiable data processing and decision-making support capabilities through trustworthy intelligent agents.
Wu Minghui, Founder, CEO, and CTO of MiningLamp Technology, emphasized that AI development has entered a new phase centered on intelligent agents. Unlike consumer-facing "one-sentence generation" applications, enterprises prioritize accuracy and traceability in production environments. DeepMiner was designed to meet the demand for transparency and reliability in enterprise scenarios.
DeepMiner adopts a Multi-Agent Architecture (MoA), enabling it to match the optimal model for processing each specific business segment, significantly enhancing system optimization efficiency compared to Mixture-of-Experts (MoE) architecture. From task decomposition and tool calling to result generation, users can transparently review each step's logic. Human-machine interaction mechanisms also allow real-time intervention to minimize "hallucinations" and ensure output verifiability.
Additionally, DeepMiner integrates commercial databases across advertising, retail, e-commerce, and other sectors to guarantee data authenticity and comprehensiveness, reducing hallucination risks from the source. Actual tests show that its hallucination rate in vertical industries significantly lower than foundation models.
Technically, DeepMiner is centrally orchestrated by an intelligent hub—the Foundation Agent—which coordinates the collaborative work of all components. The Cito model serves as the analytical decision hub, dynamically constructing specialized reasoning chains for complex business problems while narrowing action spaces via human-AI collaboration to enhance efficiency. The Mano model serves as an automated execution engine, enabling agents to truly learn to "see" and "click", achieving precise operations in complex software and browser environments. Through Foundation Agent coordination, DeepMiner merges Cito's deep reasoning capabilities with Mano's execution prowess, creating an end-to-end intelligent automation loop from business insight to operational execution.
Huang Nan, Product Manager of DeepMiner noted that the evolution from Generative AI to Agentic AI represents a paradigm shift: Agents endow large models with "eyes and hands" to proactively execute tasks. This precision in tool calling is made possible by the performance of the Browser Use Agent (BUA) and Computer Use Agent (CUA). However, foundation models aren't specifically trained for software operation, resulting in suboptimal performance in complex business scenarios.
To address this, MiningLamp Technology developed the Mano model, through continuous reinforcement learning, can autonomously explores and adapts to new platforms and workflows, demonstrating superior capabilities in real-world business scenarios. Notably, Mano has already topped two globally authoritative benchmark tests—Mind2Web and OSWorld—both achieving state-of-the-art (SOTA) performance.
Wu Minghui, Founder, CEO, and CTO of MiningLamp Technology stated: "Trustworthiness is the core standard for enterprises adopting AI, and DeepMiner's mission is to empower enterprises with trustworthy productivity by delivering trustworthy Agentic model and data."
From big data to large models, MiningLamp Technology has firmly demonstrated its ability to navigate through technological cycles, thanks to nearly two decades of deep expertise in the data intelligence. The launch of DeepMiner marks a significant strategic move into the large model era for MiningLamp Technology. Moving forward, MiningLamp Technology will also launch dedicated intelligent agents based on DeepMiner, targeting vertical industries such as finance, law, human resources, and manufacturing.
source: Mininglamp Technology
