Mininglamp Technology releases 2026 interim results: Agentic Services revenue surges 605.7% year-on-year, FDE model accelerates

Mininglamp Technology (02718.HK) officially releases its interim results for the six months ended June 30, 2026.During the reporting period, the Company generated revenue of RMB 760 million, a year-on-year increase of 18.0%; gross profit amounted to RMB 390 million, up 8.3% year-on-year; and the net loss for the period was RMB 64 million, narrowing by 68.5% year-on-year.

During the reporting period, Minglue Technology accelerated its AI Native transformation. Revenue from Agentic Services surged significantly, becoming a key driver of the Company's revenue growth. The open-source trusted Agent collaboration network, Octo, was officially launched, with a series of edge-side models being progressively deployed. Meanwhile, AI-driven organizational restructuring and FDE (Full-Dimensional Engagement) layout advanced in tandem, comprehensively rolling out strategic initiatives tailored to the new paradigm of the intelligent economy.

In the first half of the year, revenue from intelligent marketing services reached RMB 344 million, a year-on-year increase of 4.2%. The intelligent store operations system has covered more than 75,000 offline stores. The Company’s long-accumulated data capabilities and industry expertise have provided strong support for the integration of agentic services into specific business processes.

Multi-scenario deployment of agentic services has become the primary engine for revenue growth.

During the reporting period, revenue from the Company's agentic services reached RMB 171 million, a year-on-year surge of 605.7%, accounting for 22.5% of total revenue and becoming the main source of revenue growth. New customers accounted for 68%, reflecting the continued rapid expansion of the customer base.The Company applied its self-developed AI agents to strategy generation, content creation, and execution phases. This approach not only improved delivery efficiency but also expanded the customer scale, further validating the commercial viability of the performance-based payment model.

In the scenario of marketing content production,the automated production line can sustainably produce dozens of short videos meeting advertising standards per day, with the copy approval rate rising to 80%,helping brands increase the frequency of content testing and reduce production costs. New demands arising from AI search and content cognition are also being rapidly unleashed,with related services already covering over 100 leading clients across 15 industries.

The influencer marketing business is gradually shifting from single-project collaborations to platform-based operations. During the reporting period,the repurchase rate for related business clients reached 90%, and the average transaction value increased by two to three times year-over-year.As centralized procurement on platforms such as TikTok increases,influencer resources are transitioning from one-off project assets into sustainable, organizable, matchable, and reusable content supply.

The AI-native foundation continues to be strengthened, with Octo driving multi-agent collaboration into organizational workflows.

During the reporting period, Mininglamp Technology officially launched Octo, its first open-source and trustworthy agent collaboration network. Aiming to build an 'Internet of Agents,' it promotes the upgrade of AI agents from personal productivity assistants to collaborative, orchestratable, and accumulative organizational-level digital labor.

Through product concepts and functional designs such as Bot, Channel, Thread, and Loop, Octo provides agents with identity identification, task orchestration, context sharing, and work logging capabilities. This enables bots with diverse specialties to engage in distributed collaboration under human guidance, forming collective intelligence that surpasses single models through network effects.

Octo is open-sourced under the Apache License 2.0 and supports private deployment. Combined with clear data ownership design, it ensures that data sovereignty, knowledge subjects, and collaboration entities truly return to users and organizations, while protecting, amplifying, and inheriting individual tacit knowledge within respected data boundaries.

Octo efficiently connects personnel and agents within enterprises,while Octic, Mininglamp Technology's self-developed first AI-native recording hardware, further integrates high-value contexts generated from offline meetings and real-world interactions into this collaboration network, creating a closed loop and ecological synergy of 'B-side data assets + C-side product experience.'

Currently, Octo covers web platforms, desktop clients, mobile devices (iOS and Android), browser plugins, and command-line interfaces (CLI), achieving seamless cross-platform interoperability.The Octo platform demonstrates stable performance in high-concurrency tasks and complex environments involving multiple members and agents. Within the company, over 1,800 employees and 5,100 agents are already collaborating seamlessly through Octo.Furthermore, the company is closely engaging with major clients across various industries to advance the deployment of Octo in scenarios such as content production and organizational collaboration.

Looking ahead, Octo will continue to embrace an open-source philosophy, co-creating new AI-native collaborative models with developers, enterprise clients, and ecosystem partners. This aims to drive the implementation of trustworthy, controllable, and sustainable private Agentic AI in real-world work scenarios.

Technology-driven enhancement of capabilities for general-purpose on-device model series.

As large language models penetrate deeper into enterprise scenarios, issues associated with cloud deployment—such as data security risks, ongoing inference costs, and response latency—are becoming increasingly prominent. Consequently, on-device deployment is emerging as a key development direction for the industry.

Based on in-depth exploration of the inference mechanisms of Mixture of Experts (MoE) architectures, the company has proposed a new paradigm for model construction known as 'Scaling Out.'Models built using the Scaling Out approach achieve performance comparable to frontier models in target capability domains and significantly outperform distilled models of similar parameter scales, striking an elegant balance between model size and accuracy.

Leveraging this methodology,Minglue Technology has developed a range of general-purpose on-device models. These cover core capabilities including multimodal understanding (GUI visual-language-action, image-text, and speech); agent operations (code generation, tool invocation, and in-context learning); and general cognitive reasoning.The relevant models can combine capabilities from different domains based on specific customer scenario requirements and flexibly customize model sizes.

Furthermore, the company has independently developed and open-sourced Cider, an edge-side inference acceleration SDK based on the Apple MLX ecosystem.This SDK is deeply optimized for macOS and Apple Silicon platforms, addressing gaps in the native MLX framework regarding activation quantization and specific tensor computations. By enhancing edge-side inference efficiency, Cider lowers the barrier to deploying edge models, providing an engineering foundation for their operation on personal computers and other edge devices.

Continuously building an AI-native organization, with the FDE model accelerating entry into customer sites.

The company is upgrading AI from a standalone tool to a foundational productivity layer permeating the entire organization, advancing AI-native transformation along three main lines: organizational efficiency enhancement, service empowerment, and capability assetization.During the reporting period, the company established AI-driven efficiency quantification baselines in 172 report delivery scenarios, achieving a comprehensive efficiency improvement of 56.1% and accumulating 359 reusable Skills., enabling individual experience to be transformed into organizational assets through Skills and Bots.Certain business lines saw a 40% year-over-year decrease in delivery labor costs while simultaneously achieving a 40% year-over-year increase in service volume.

For enterprise AI to truly integrate into business operations, it requires personnel who can understand customers' not-yet-fully-defined needs, organize industry expertise, business processes, and data permissions, and then embed Agents into actual workflows.During the reporting period, approximately 30% of the company's internal delivery staff transitioned into Frontend Deployment Engineers (FDEs).With support from Octo, FDE collaborates with clients to clarify requirements, translate business objectives into concrete workflows, and orchestrate models, data, and functional agents to embed enterprise AI directly into client operations.

Furthermore, the company plans to leverage Pulian Software’s (300996.SZ) existing product capabilities, client scenarios, and implementation experience. Driven by the dual engines of 'AI + Information Technology Innovation,' it aims to deepen integration into core ERP systems and broader business processes of large enterprises. By enabling intelligence across all scenarios, the company seeks to expand into a vast incremental market characterized by high customer stickiness and high average revenue per user.Meanwhile, the company will facilitate the transition of the Pulian Software team to the FDE model, enhancing product and service capabilities that align closely with client business scenarios to support the intelligent transformation of large corporate groups.

Mininglamp Technology is continuously validating the industrial value of trustworthy Agentic AI through real-world business closed loops. Aligning with its long-term vision of 'building organizational intelligence in the AI era,' the company will continue to deliver results in real tasks and scale up applications, transforming expert knowledge, production processes, and collaborative relationships into accumulable and reusable organizational capabilities.

$MININGLAMP-W (02718.HK)$

29K Views