In the ever-evolving and competitive landscape of artificial intelligence, one company stands out as particularly unique: Kunlun WanweiUnlike giants such as Alibaba and Baidu, which boast market valuations in the hundreds of billions, Kunlun Wanwei is considered a mid-sized entityNevertheless, it has managed to carve out a significant niche in the resource-intensive AI sector, establishing itself as a notable player amidst skepticism over its perceived "copycat" approach to AITwo years have passed since its foray into the field, and Kunlun Wanwei has demonstrated persistence that belies initial doubts.

On January 6 of this year, Kunlun Wanwei launched its "Tiangong Model 4.0," a leap forward in Chinese logical reasoning capabilities, directly competing with OpenAI's offeringsThis introduction marked a significant milestone as the first Chinese model to effectively handle queries with end-to-end speech processing, placing it in competition with seasoned models like GPT-4.

What enabled Kunlun Wanwei to secure its place at the proverbial table? Founder Zhou Yahui attributes its entry into the AI arena to the timeliness of its initiatives

"If we hadn't started early in this large model game, we wouldn't be where we are today," he explained, highlighting the importance of timing and commitment in the AI sectorHowever, as the initial advantage of being an early player begins to wane, Kunlun Wanwei is now pivoting towards a new focus: the application of AI, which it perceives as a more stable and promising domain.

Developing a Super App requires not only technological proficiency but also innovative product thinkingAs the momentum of scaling laws slows down, Kunlun Wanwei is attempting to leverage its operational expertise from the Internet era, seeking to endure longer in the increasingly crowded market of large models.

The question now is: will this strategy be successful?

Racing Ahead in AI

How did Zhou Yahui, who graded ByteDance's AI strategy a 'C' just eighteen months ago, formulate Kunlun Wanwei's approach? It appears the answer hinges on speed.

This current wave of large models is still in its infancy

• Energy Crisis Meets Fed's Monetary Tightening
• Global Stock Markets Surge: Will A-Shares Follow?
• U.S. Inflation Nearing Its Peak
• Kunlun Wanwei's AI Ambitions and Hurdles
• Battery Firm Targets $1B Off-Road EV Market

In a conversation last October, CEO Fang Han remarked, "Kunlun Wanwei is experimenting in multiple directions quickly, embracing the ethos of fast failure and rapid shifts in focus." Throughout the transformative period of "All in AGI and AIGC" spanning over two years, the pace at which Kunlun Wanwei executed its plans played a crucial role in securing its status within the AI industry.

In 2023, the company seized a first-mover advantage in large models, being one of the earliest players in what has been termed the “Battle of the Hundred Models.” This move proved rewarding as its stock price soared 230% following the launch of a GPT-3.5 equivalent model named "Tiangong," propelling the company's valuation to heights of 79.3 billion yuan.

As the user base for Kunlun Wanwei's "Tiangong AI" continues to grow, it is recognized as one of the few applications in China to surpass five million active users monthly

The app stands out for pioneering AI search functions within the domestic landscapeEmployees have disclosed that Tiangong AI was also the first among peers to launch in terms of advertising initiatives, capitalizing on a window of opportunity.

In the AI search sector, the company has been able to secure significant initial benefits, rapidly acquiring users due to its timely product deploymentIncidentally, recent statistics from January to October of 2024 reflected that "Tiangong AI" achieved a remarkable cumulative download of 11.7 million, solidifying its competitive edge in a burgeoning market.

However, the question remains as to whether the early advantages that allowed Kunlun Wanwei to establish its foothold will ensure its continued presence at the forefrontFostering improvements in foundational model capabilities is paramount, especially as the company endeavors to keep pace with leading players while reinforcing its own AI application framework.

In November, Kunlun Wanwei introduced "Tiangong Skywork O1," a direct competitor to OpenAI's offerings, coupled with releases of an open-source model based on the Llama 3.1 architecture

The company is confident these advancements, driven by enhancements like Q* algorithms and reinforced learning capabilities, lead to superior performance in mathematical and coding tasks, claiming to experience a state-of-the-art capability surpassing Alibaba’s Qwen-2.5-7B model.

Moreover, in the month of launch, the company stepped up its efforts in developing multimodal capabilities, with the debut of the Skywork 4.0 modelThis model demonstrates efficient response times and supports real-time interactions, emotional understanding, personalized memory, and customizable tones in its functioning.

Nevertheless, the positioning Kunlun Wanwei once enjoyed in the foundational large model market is becoming less distinct as more companies in China make headway into multimodal functionsFor instance, several enterprises, including Zhizhu AI, iFlytek, and West Lake Cognitive Technologies, launched their end-to-end speech models around September, putting Kunlun Wanwei at a disadvantage as it trailed by nearly three months.

The current landscape of Kunlun Wanwei with respect to GPT-4o-like products can be interpreted as a reflection of its broader situation: with competition intensifying and players accelerating their advancements, Kunlun Wanwei finds its once-assertive pace diminishing.

As development in foundational large models decelerates, the company is striving to utilize AI products as a means of regaining competitive edges

Nonetheless, Kunlun Wanwei faces significant hurdles stemming from the financial clout and resource allocations of larger companiesThis has presented a considerable barrier for smaller contenders like itself.

In the arena of large models, Kunlun Wanwei has encountered challenges synonymous with its designation as a mid-sized venture, encountering difficulties competing against both established conglomerates and startup innovatorsIn terms of multimodal enhancements, the latest variant of its large model is the Tiangong Model 3.0, which launched in April and boasts 400 billion parameters, claiming to surpass Grok’s capabilities, thereby marking it as the largest open-source MoE model at the time.

However, the rapid iterations of multimodal capabilities exhibited by other Chinese companies have outpaced it significantlyJust days ago, Deepseek launched its V3 model, which surpassed many baseline metrics compared to top-tier closed models such as GPT-4o.

It becomes apparent that as improvements in scaling laws diminish, the pretraining efficacy of foundational models is constrained

Exclusive reliance on pretraining is no longer sufficient to ensure rapid enhancements in model abilities.

Reflecting on the trajectory of the industry, Li Kaifu has noted that the cost and implications of training larger models would be substantially prohibitive for startupsIn this changing arena, the diminishing returns associated with large model advancements are beginning to surface, as Fang Han has indicated during an AI conference, suggesting that the cycle of advantage from state-of-the-art (SOTA) technologies is becoming increasingly fleeting.

With that said, the alternative path Kunlun Wanwei explored now appears to broaden its scope significantlyZhang Fang has remarked, "The innovation in product and business models focused on AI products is lagging compared to technological advancementsYet, for end-users, innovation in products and business models carries greater weight." Over the span of two years, the company has rapidly iterated its offerings across five key areas: AI short films, AI social networking, AI search, AI music, and AI gaming.

These platforms can essentially be streamlined into two main business lines: AI search and AI UGC platforms

Fang asserts that AI search will emerge as a robust Super App for enhancing user productivity, whereas the AI UGC platform provides a space for user expression, thereby fostering connections between users and applications.

Moreover, the company has substantial backing in terms of product and operational expertise trades through its acquisition of Opera and initiating its overseas endeavors, allowing it to harness growth potential in international markets.

The underlying advantages of Kunlun Wanwei are becoming evident in its product dataNotably, it maintains opportunities to compete with big players and retain relevance in AI search and social applicationsFor instance, Tiangong AI boasts over five million monthly active users, while its social product Linky surpassed 3.78 million international monthly active users, ranking sixth among outbound AI products.

Despite success, Kunlun Wanwei faces challenges within its application segments, much like the broader complications associated with growth in large model enterprises

For example, while its self-constructed indexing library theoretically provides an edge over many competitors, the larger companies with established browser products and enriched datasets can effortlessly consolidate user bases.

Take, for example, the surge in traction for Baidu's "AI Search." In December, the product's web access witnessed a monthly increase of over 90%, hitting above 11 million visits within a single month, significantly outpacing Kunlun Wanwei's Tiangong AI.

With Baidu now fully integrated its AI search functionalities onto its homepage and capturing traffic through established routes, the timeline grows shorter for players like Kunlun Wanwei to stake their claims in a competitive marketplace.

Kunlun Wanwei: Clinging to the AI Lifeline

Upon inspecting Kunlun Wanwei's strategic transitions since its inception, it appears the company possesses an uncanny ability to identify and leverage pivotal trends effectively

Having initially launched as a gaming entity, it diversified by acquiring a browser and aggressively ventured into international marketsIn 2010, Kunlun Wanwei made its global aspirations clear by acquiring various applications such as Grindr and Opera, positioning itself strategically in overseas markets.

For now, AI has materialized as Kunlun Wanwei's third trump card"As a mid-sized internet company, our sole ambition is to evolve into a larger company, hence we constantly seek out secondary curves," Fang states regarding the impetus behind embracing AI.

Success would undoubtedly elevate Kunlun Wanwei's status in the industry, enabling it to ascend to the ranks of larger counterpartsConversely, failure would likely see it stabilize back into the middle tier of the market, continuing the search for the next viable trend to pursue.

Critics of Kunlun Wanwei often assert that it acts opportunistically to capitalize on trending technologies

However, the current trajectory leads it to a more accessible and relatable solution than previous focuses such as the metaverse or renewable energy—AI remains a tangible asset for the company.

It's precisely because of this that Kunlun Wanwei boldly proclaims, "All in AGI and AIGC," allocating vast resources toward AIIn its most recent quarterly report, the company reported a 84.47% increase in R&D expenses to 1.144 billion yuan, alongside an impressive 800 million yuan investment in chip acquisitions during 2023 alone to ensure sufficient computational reserves.

For Fang, these expenses are a "necessary journey," with expectations that technological evolution will eventually lead to reduced losses moving forwardHe remains optimistic that initial investments will yield favorable market share returns as downturns in computational costs materialize.

Many securities firms have rated Kunlun Wanwei as a "buy" or "hold." Huatai Securities, in a recent report, underscored its competitive edge in both large models and applications, predicting continuous positive developments within its application pipeline.

Looking ahead to 2025, industry insiders are betting that this year will mark an explosive growth period for AI applications, with reduced operational costs enabling companies to buoy themselves through financial independence.

Yet from the outset of its journey toward AGI, it’s essential for all domestic companies, including Kunlun Wanwei, to grapple with commercialization concerns

If the company can swiftly navigate the complexities of its constraints, testing and identifying a feasible Super App, it would deem any current losses a worthwhile investment.

On the contrary, neglecting a previously lucrative opportunity could lead to substantial ramifications, as Fang suggestsFor Kunlun Wanwei, leaving the AIGC landscape isn't merely about financial expenditures; it instead revolves around maintaining relevance against its counterparts.

Having invested countless resources to establish a multimodal foundation, Kunlun Wanwei is poised to continue marketing its existing AI search and UGC initiatives, while also unveiling a series of innovative applications to convert initial investments into successful products.

Fang's outlook on AIGC and AGI encapsulates a mindset that mirrors all domestic AI firms: "Reaching AGI is certainly our aimBut to simply survive, AIGC is our means." This summation underscores the dual pursuit of ambition and practicality that drives Kunlun Wanwei forward in uncertain times.