As the industry forges ahead into uncharted territories, especially amidst escalating computational expenses and a growing appetite for data privacy, the quest for a fresh breakthrough in large model development has become a shared goal across the sector. This necessitates that large model participants not only boast cutting-edge algorithms but also possess industry-integrated device terminals to sidestep the prevailing dilemmas and hurdles in large model progression.

An increasing number of players are intensifying their R&D endeavors in edge large models, striving to overcome the current bottlenecks in large model evolution. Against this backdrop, we observe edge large model innovators garnering attention from capital markets and edge large models emerging as a relatively definitive new avenue for growth.

As edge large models carve out a new and highly promising path, particularly with financial backers beginning to zero in on this trajectory, it is clear that competition in edge large models is poised to enter a fresh chapter in the near term. For every participant aiming to leave a mark in this arena, discovering a means to genuinely establish edge large models as a pivotal sector within the industry is paramount for triumph in the new phase of large model evolution.

Capital Markets Embark on a Deep Dive into the Edge Large Model Arena

According to a report issued by the China Industrial Internet Research Institute, edge large models are poised to become a new driving force for AI growth. Investors, exemplified by Hongtai Fund, assert that 'efficiency' is the cardinal principle of large models. Edge large models, being closer to users and offering enhanced privacy, have evolved into high-knowledge-density application scenarios. As a pioneer in edge large models, Minwall Intelligence's trajectory exemplifies this trend. Minwall Intelligence's MiniCPM series models (e.g., MiniCPM-o 2.6) boast a compact parameter size (e.g., 8 billion) yet support full-modal interaction, with over 10 million downloads, and have been deployed in automotive, judicial, education, and other sectors. Since April 2024, Minwall Intelligence has secured four rounds of financing. This data underscores that capital markets are displaying keen interest in edge large models and are initiating a new wave of strategic positioning.

A pivotal reason for the capital market's fervent interest in edge large models is their proximity to users. By bringing large models closer to the end-user, edge large models can tackle the disconnect and challenges between large models and users. For the large model sector, which has ventured into deeper waters, leveraging edge large models as a catalyst to uncover new growth prospects is emerging as a novel highlight in participants' strategies.

Beyond their user proximity, another significant factor driving the capital market's robust investment interest in edge large models lies in the core of artificial intelligence, which transcends mere size (large vs. small) or location (edge vs. cloud), but rather hinges on the distinction between high and low knowledge density. For any participant striving to make a mark in the new phase of large model competition, bolstering knowledge density through edge large models to achieve a redistribution of traffic and industry scenarios has become a new objective and direction.

By positioning intelligent terminals nearest to users, fostering genuine user adoption, and addressing the challenges of high computational costs and privacy demands in large model development, a new objective and direction have surfaced. It can be stated that command over edge large models will dictate success or failure in the profound waters of large model development. This is why we witness the capital market concentrating its investments on edge large models, which are closest to users.

Competition in the Large Model Arena Ushers in a Fresh Chapter

With the gradual deployment and application of deepseek, AI competition globally has escalated to a fierce stage. A parallel scenario is unfolding in the large model sector. By synthesizing and defining the trajectory of the large model sector, it is evident that competition is now more centered on quality and efficiency, rather than the prior emphasis on first-mover and diversified advantages. In this context, how to deeply meld large models with industry scenarios and genuinely connect with users, industries, and scenarios has become a new development breakthrough. Against this backdrop, we observe new directions, typified by edge large models, emerging as prominent indicators of the large model sector entering a fresh phase.

A key reason for edge large models becoming a new direction is that they are not merely distant from users and industries but have descended to the frontlines, enabling seamless integration with users and industries. Through this approach, edge large models can not only resolve the issues stemming from the vast and comprehensive nature of traditional large models but also facilitate genuine implementation in real-world scenarios and industries, forging close ties with users.

It can be argued that edge large models have resolved the input-output dilemma in large models, uncovering a novel solution for large model development.

Moreover, edge large models can craft large models with superior performance, lower costs, reduced power consumption, and accelerated speeds under equivalent parameters. Through this method, large models have discovered a new development path entirely distinct from the past, entering a novel stage that diverges from traditional development models. Hence, edge large models, characterized by their lightweight nature and efficiency, are emerging as a new and definitive direction of development.

Full Commercialization Emerges as a New Frontier

As a plethora of large models commence implementation and application across diverse industries and scenarios, particularly as large models enter a new phase of competition, the true litmus test for large model participants is no longer merely about possessing what others lack but increasingly about their ability to monetize. Therefore, for the current large model sector, how to swiftly and fully achieve commercialization and establish a sustainable business closed loop has become a new challenge for large model participants.

At this juncture, how participants can bring their large models closer to users and industries has become pivotal in establishing new traffic and industry patterns. Against this backdrop, we observe new directions, typified by edge large models, capable of rapid commercialization, becoming the development focus for an increasing number of participants.

Taking Minwall Intelligence as an exemplar again, in January of this year, Minwall Intelligence unveiled the edge full-modal model - Minwall Cannon MiniCPM-o 2.6, which leads the industry in pivotal capabilities such as 'continuous viewing, real-time listening, and natural speaking,' attaining international advanced levels in overall capability. Previously, the Minwall Cannon MiniCPM garnered global acclaim for its ability to achieve more with less, being efficient and cost-effective, successively realizing the functions of models such as ChatGPT, GPT-4V, and GPT-4o on the edge. In 2024, it ranked among the top in China for download volume and popularity on Hugging Face. As of now, the cumulative download volume of the Minwall Cannon MiniCPM series across all platforms has surpassed 10 million.

Everything indicates that large model sectors typified by edge large models, which are more conducive to commercialization, are becoming the new development direction. For every participant striving to make a difference at this stage, how to integrate high-performance edge AI into people's production and lives and discover a means to achieve rapid commercialization is crucial for their success in this entirely novel phase. When the input and output of large models achieve seamless coordination and a relatively balanced relationship is established between productization and commercialization, we can truly uncover a healthy and sustainable new development path suitable for large models.

Epilogue

As large model development ventures into deeper waters, particularly as AI competition globally intensifies, developing large models in a more lightweight, efficient, and commercializable manner has become a new consensus. Against this backdrop, we observe directions typified by edge large models becoming the pivotal tracks for determining success in the new phase of the industry. With the full-fledged support of capital markets, the exploration of large model participants, and the acceleration of commercialization, edge large models are showcasing prominent advantages. Uncovering more sustainable and enduring new paths for the large model sector, with edge large models as the definitive direction, is a question that every participant must answer.