Listen To This Article

Listen to this post

Ready to play

The Rise of Chinese Large Language AI Models: A Comprehensive Analysis

The Rise of Chinese Large Language AI Models: A Comprehensive Analysis

The Rise of Chinese Large Language AI Models: A Comprehensive Analysis of Recent Advancements

By Brian Clark | Published: [Insert Date, e.g., April 15, 2025]

1. Executive Summary

The landscape of artificial intelligence is undergoing a significant transformation with the rapid evolution of large language models (LLMs). In this dynamic environment, China has emerged as a formidable force, demonstrating remarkable progress in the development and deployment of advanced LLMs. This report provides a comprehensive analysis of the recent advancements in Chinese LLMs, primarily focusing on developments in 2024 and early 2025. Key players such as DeepSeek, Alibaba, Tencent, ByteDance, Baidu, and Hon Hai have unveiled groundbreaking models showcasing enhanced capabilities in areas like reasoning, multilingualism, multimodality, and cost-efficiency. A notable trend is the increasing prevalence of open-source models, exemplified by DeepSeek's R1, which challenges the dominance of proprietary models. These advancements are not occurring in a vacuum; they are shaped by a complex regulatory framework and evolving ethical considerations specific to China. The potential impact of these new Chinese LLMs spans across numerous industries, from finance and healthcare to education and manufacturing, signaling a significant shift in the global AI landscape and technological power balance. Strategic collaborations and partnerships within the Chinese AI ecosystem are further accelerating this progress, positioning China as a major contender in the ongoing AI revolution.

2. Introduction

The advent of artificial intelligence, particularly in the realm of Large Language Models, has ushered in an era of unprecedented technological possibilities. These sophisticated models, capable of understanding and generating human-like text, are at the forefront of numerous innovations across various sectors. Recognizing the transformative potential of AI, China has set ambitious goals to establish itself as a global leader in this field, making substantial investments in research and development (2). This commitment has fostered the rapid growth of a robust Chinese AI ecosystem, increasingly capable of challenging the dominance of established global players. This report focuses on the most recent developments in Chinese LLMs, with a particular emphasis on models announced and discussed in 2024 and the early months of 2025.

The significance of these advancements extends beyond mere technological progress. Chinese LLMs are demonstrating a growing ability to compete with leading global models originating from the United States (5). This competition has profound implications, not only for the development of localized AI solutions tailored to the unique needs of the Chinese market but also for the broader global AI landscape and the shifting balance of technological power (1). Understanding China's progress in this critical technology is therefore essential for comprehending the future trajectory of AI on a global scale. This report will delve into the specifics of these recent LLM developments, examining their technical characteristics, reported capabilities, performance benchmarks, the regulatory and ethical context in which they are emerging, their potential impact across industries and society, and the collaborations that are driving this innovation.

The past year and early 2025 have witnessed a flurry of announcements of new and significantly upgraded large language models from various Chinese organizations, signaling an intense period of innovation and competition.

3.1 DeepSeek's Breakthrough

A particularly noteworthy development has been the emergence of DeepSeek, a Chinese AI company that has rapidly gained prominence (9). In December 2024, DeepSeek announced its new general-purpose large language model, DeepSeek-V3, followed swiftly in January 2025 by DeepSeek-R1, an AI model specifically designed for complex logical tasks (7, 13). These models have garnered significant attention for their reported cost-efficiency and performance that rivals or even surpasses that of models from OpenAI (10). DeepSeek-R1 stands out for its strong focus on reasoning capabilities and its availability as an open-source model, a strategic move that has been lauded within the AI community (14, 15). The ability of DeepSeek to achieve such high performance with significantly lower training costs, reportedly around $5.6 million for DeepSeek-V3, compared to the much larger investments by US-based companies, suggests a potential paradigm shift in how advanced AI models can be developed (5).

3.2 Alibaba's Qwen Series

Another major player in the Chinese LLM landscape is Alibaba, which has continued to upgrade its Qwen series of models (61, 62). In January 2025, Alibaba released an upgraded version of its AI model, Qwen 2.5 Max, claiming that it "comprehensively outperformed" its Chinese competitor DeepSeek's model in certain benchmark tests and also reportedly outmatched LLMs from OpenAI and Meta (17). Alibaba has also introduced specialized models within the Qwen 2.5 family, such as Qwen2.5-Coder, built for coding tasks, and Qwen2.5-Math, optimized for mathematics (16). Furthermore, Alibaba Cloud's Qwen-2.5-1M has been highlighted for its ability to handle extremely long questions and engage in longer and more in-depth conversations, indicating a focus on improving contextual understanding (64). The rapid iteration and specialization within Alibaba's Qwen series underscore the intense competition among Chinese tech giants to establish themselves as leaders in the LLM domain.

3.3 Tencent and ByteDance's LLMs

Following the impressive launch of DeepSeek-R1, other Chinese tech titans, including Tencent and ByteDance, also announced their own large language models, with both companies asserting that their models exceeded the capabilities of DeepSeek-R1 (3). ByteDance, the parent company of TikTok, released Doubao 1.5 Pro in January 2025 (18, 21). This model emphasizes a "resource-efficient" training approach that does not compromise performance, achieving high levels of proficiency in knowledge retention, coding, reasoning, and Chinese language processing (4, 20). ByteDance highlighted the integrated train-inference design of Doubao 1.5 Pro, which balances optimal performance with efficient inferencing costs (19, 22, 23, 24, 25). The swift announcements from Tencent and ByteDance following DeepSeek's breakthrough demonstrate the highly competitive nature of the Chinese LLM market and the urgency to keep pace with emerging advancements.

3.4 Baidu's Ernie Bot Advancements

Baidu, a long-established leader in China's AI sector, has also made significant strides with its Ernie Bot (27, 29). In March 2025, Baidu unveiled the latest versions of its models, Ernie 4.5 and Ernie X1, claiming that these models outperform OpenAI's GPT-4.5 in several benchmark tests while costing only a fraction to use (28, 31). Ernie X1 is presented as Baidu's first multimodal reasoning model, capable of handling complex problem-solving tasks (30), while Ernie 4.5 features enhanced multimodal capabilities, allowing it to understand and integrate various data types, including text, images, audio, and video (26). Baidu's continued innovation with the Ernie Bot series, with a strong emphasis on multimodal functionality and cost-effectiveness, positions it as a key competitor in the Chinese AI landscape (63, 90).

3.5 Hon Hai Research Institute's FoxBrain

In March 2025, Hon Hai Research Institute, backed by Hon Hai Technology Group (Foxconn), announced the launch of its first Traditional Chinese Large Language Model, named FoxBrain (32, 33, 35, 36). This development marks a significant milestone in Taiwan's AI technology, as the model was trained using an efficient and lower-cost method completed in just four weeks (32). Initially designed for internal applications within the Group, covering functions such as data analysis, decision support, and code generation, Hon Hai has plans to open-source FoxBrain and collaborate with technology partners to expand its applications (32, 34). FoxBrain has demonstrated strong comprehension and reasoning abilities, particularly excelling in mathematical and logical reasoning tests compared to other Traditional Chinese models and even surpassing Meta's models of the same class in mathematical reasoning ability (32, 37, 38). Hon Hai's entry into the LLM arena with a focus on the specific linguistic needs of Traditional Chinese users underscores the importance of tailored AI solutions.

3.6 Other Notable Models

Beyond these major announcements, other significant models have emerged from the Chinese AI ecosystem (8, 48, 49). Moonshot AI's Kimi k1.5 has garnered attention for reportedly matching or even outperforming OpenAI's o1 model, particularly in areas like mathematics, coding, and the ability to understand both text and visual inputs (4, 39, 40, 41, 42, 43, 44, 45, 46, 47). Zhipu AI's GLM-4 Plus is recognized for its capabilities in processing long-form text, following complex instructions, and handling multimodal interactions (5, 50, 51, 53, 57). MiniMax has also continued to advance its multimodal LLMs, further contributing to the vibrant and competitive nature of the Chinese AI landscape (5). The emergence of these numerous AI unicorns and startups alongside established tech giants highlights a dynamic innovation ecosystem in China's pursuit of advanced LLM capabilities.

The recent wave of Chinese LLMs showcases a variety of technical specifications and architectural innovations aimed at enhancing performance, efficiency, and specific capabilities.

4.1 Mixture-of-Experts (MoE) Architecture

A notable architectural trend among the new Chinese LLMs is the adoption of the Mixture-of-Experts (MoE) approach. DeepSeek-R1 and its predecessor DeepSeek-V3 both utilize an MoE architecture with a substantial number of total parameters (671 billion), but only a fraction of these parameters (37 billion) are activated for each input token (14, 58). This selective activation allows the models to achieve high levels of performance while maintaining computational efficiency. Similarly, ByteDance's Doubao 1.5 Pro is built on a sparse MoE architecture, enabling it to deliver performance comparable to a much larger dense model with a significantly smaller number of activated parameters (18). This strategic use of MoE architectures appears to be a key factor in enabling Chinese LLMs to compete effectively with global leaders in terms of both performance and cost-effectiveness.

4.2 Training Data and Methodologies

The development of these advanced LLMs relies on massive datasets and sophisticated training methodologies. DeepSeek-V3 was trained on an extensive dataset comprising 14.8 trillion tokens, utilizing mixed FP8 precision training on a cluster of 2048 H800 GPUs (58). DeepSeek-R1 further built upon this foundation through large-scale reinforcement learning, specifically focusing on enhancing its reasoning abilities (59). Hon Hai's FoxBrain employed a highly efficient training strategy, completing its pre-training in approximately four weeks using 120 NVIDIA H100 GPUs. This process involved proprietary data augmentation methods and a unique Adaptive Reasoning Reflection technique to train the model in autonomous reasoning (32). Alibaba's Qwen2.5 series benefited from an expanded training dataset, growing to 18 trillion tokens, with significant improvements in data quality and the training process (60). ByteDance's Doubao 1.5 Pro was trained using a "resource-efficient" approach, incorporating an integrated design that optimizes both the training and inference phases (19). Moonshot AI's Kimi k1.5 emphasized long-context scaling and reinforcement learning optimization in its training (39, 41). Zhipu AI's GLM-4 Plus was pre-trained on a vast corpus of ten trillion tokens, primarily in Chinese and English (50). The diversity of these training datasets and methodologies underscores the intense research and development efforts underway in China to push the boundaries of LLM capabilities.

4.3 Key Technical Features

Beyond the core architectures and training approaches, several key technical features contribute to the advanced capabilities of these Chinese LLMs. DeepSeek-V3 incorporates Multi-Head Latent Attention, which enhances memory efficiency, and Multi-Token Prediction, which significantly increases inference speed (9). Hon Hai's FoxBrain, based on the Meta Llama 3.1 architecture with 70 billion parameters, boasts an impressive context window length of 128,000 tokens (32). ByteDance's Doubao 1.5 Pro supports extended context windows ranging from 32,000 to 256,000 tokens, enabling more effective processing of long-form text (19). Moonshot AI's Kimi k1.5 also features an enhanced context window of 128,000 tokens (39). Zhipu AI's GLM-4 Plus offers a context window of 128,000 tokens, with a specialized version, GLM-4-Long, extending this to an exceptional 1 million tokens (52). These advanced technical features highlight the focus on improving efficiency, handling extensive contextual information, and catering to a wide range of application requirements.

Table 1: Technical Specifications of Prominent New Chinese LLMs (Early 2025)
Model Name Developer Release Date Parameters (B) Architecture Training Data (Trillions of Tokens) Context Window (Tokens)
DeepSeek-R1 DeepSeek Jan 2025 671 (Total) MoE 14.8 128,000
DeepSeek-V3 DeepSeek Dec 2024 671 (Total) MoE 14.8 128,000
Qwen 2.5 Max Alibaba Jan 2025 72 Transformer 18 128,000 (up to)
Doubao 1.5 Pro ByteDance Jan 2025 140 (Effective) MoE Not Specified 32,000 - 256,000
Ernie 4.5 Baidu Mar 2025 Not Specified Proprietary Massive Not Specified
Ernie X1 Baidu Mar 2025 Not Specified Proprietary Massive Not Specified
FoxBrain Hon Hai Research Inst. Mar 2025 70 Llama 3.1 0.098 (B tokens) 128,000
Kimi k1.5 Moonshot AI Jan 2025 Not Specified Transformer Diverse Multimodal 128,000
GLM-4 Plus Zhipu AI Oct 2024 Not Specified GLM 10 128,000
GLM-4-Long Zhipu AI Oct 2024 Not Specified GLM 10 1,000,000
Note: Parameter counts for MoE models often refer to the total number of parameters, with only a fraction being active at any given time.

The newly unveiled Chinese LLMs exhibit a broad range of capabilities, leading to diverse potential applications across numerous industries. These models demonstrate strong proficiency in fundamental natural language processing tasks, including understanding and generating human-like text (6). Many excel in more complex cognitive tasks such as mathematics, coding, and logical reasoning (7). Furthermore, a significant number of these models offer multilingual support, with a primary focus on Chinese and English, and some extending to nearly 30 different languages (1). The integration of multimodal capabilities, allowing these LLMs to process and generate content across text, images, audio, and video, further expands their potential applications (4).

The potential applications of these advanced LLMs are vast and span across various sectors. In the finance industry, they can be leveraged for tasks such as fraud detection, algorithmic trading, risk assessment, customer support automation, and sophisticated financial data analysis and reporting (65, 66, 67, 68, 69, 70). The healthcare sector can benefit from LLMs in areas like clinical decision support, enhanced patient interaction, medical research, accelerating drug discovery, analyzing extensive medical literature and patient health records, and even in the interpretation of medical imaging (65, 72, 73, 74, 75, 76). In education, these models hold the promise of personalized learning experiences, acting as AI tutoring systems, aiding in content creation, automating assessment processes, and providing in-depth analysis of student learning patterns (65, 78, 79, 80, 81). The manufacturing industry can see significant impact through applications in smart manufacturing processes, supply chain optimization, quality control mechanisms, and robotic intervention in production lines (32). The e-commerce and retail sectors can utilize LLMs to enhance product recommendations, automate marketing content generation, deploy sophisticated chatbots for customer service, and develop highly personalized product recommendation engines (65). The realm of content creation and media stands to be transformed by the ability of these models to automate content generation, create engaging video content, and even produce tailored advertising scripts (26). In the government and public services domain, LLMs can aid in data analysis, provide decision support, assist in cataloging vast amounts of information, and potentially contribute to the development of smart city initiatives and enhanced public safety measures (26). The transportation and logistics industries can leverage these models for advancements in autonomous driving technologies, route optimization algorithms, and overall logistics automation (65). The legal profession can benefit from the ability of LLMs to efficiently review documents, analyze contracts, and conduct comprehensive legal research (65). Finally, in customer service, AI-powered virtual assistants and chatbots powered by these LLMs can provide round-the-clock support and handle a wide range of customer inquiries (65).

An emerging trend is the development of sophisticated AI agents built upon these LLMs. These agents are capable of performing deep research and executing practical operations, exemplified by Zhipu AI's AutoGLM (84, 55, 87, 88, 89). Similarly, Monica's Manus is reported to possess the ability to make decisions and execute tasks autonomously with minimal prompting (84). Baidu's Ernie 4.0 has even demonstrated the capacity to function as a comprehensive advertising team and tackle complex problem-solving scenarios (26). The increasing sophistication of these AI agents signifies a move towards more autonomous and proactive AI systems with the potential to significantly impact various aspects of work and daily life.

A critical aspect of evaluating the advancements in Chinese LLMs is their performance relative to globally recognized leading models, primarily those developed in the United States. Recent reports and benchmark results offer valuable insights into this comparative landscape.

DeepSeek's R1 model has been reported to perform at a level comparable to or even exceeding that of OpenAI's GPT-4, Meta's Llama 3.1, and Anthropic's Claude, while notably being more cost-efficient to train and operate (3, 11). Furthermore, DeepSeek-V3 has been shown to match the capabilities of OpenAI's GPT-4o and Anthropic's Claude Sonnet 3.5 in certain benchmark tests (10). Notably, DeepSeek-R1-Distill-Qwen-32B, a distilled version of DeepSeek-R1, has outperformed OpenAI's o1-mini across various benchmarks, achieving new state-of-the-art results for dense models (14). These comparisons suggest that DeepSeek has emerged as a strong contender in the global LLM race, particularly with its open-source approach and emphasis on efficiency (91).

Alibaba has also made claims of superior performance with its Qwen series. The upgraded Qwen 2.5 Max was reported to comprehensively outperform DeepSeek's models in specific benchmark tests and also to surpass LLMs from OpenAI and Meta (17). The Qwen 2.5 series has gained significant traction within the developer community, evidenced by over 40 million downloads from platforms like Hugging Face and ModelScope (16). Moonshot AI's Kimi k1.5 is also reported to match or outperform OpenAI's o1 model in crucial areas such as mathematics, coding, and the understanding of multimodal inputs (4, 42, 44, 47). These competitive claims and adoption rates highlight the intense rivalry and rapid progress within the Chinese LLM ecosystem.

Baidu's Ernie Bot series has also been benchmarked against leading global models. Baidu claimed that Ernie 4.0 rivaled the capabilities of ChatGPT-4 upon its release in October 2023 (4, 26). More recently, Ernie 4.5 has reportedly outperformed OpenAI's GPT-4.5 in multiple multimodal benchmark evaluations (28). Baidu has also asserted that its new reasoning model, Ernie X1, delivers performance comparable to DeepSeek R1 but at half the price point (31). These claims indicate Baidu's continued efforts to position its Ernie models as highly competitive alternatives to Western counterparts.

Overall, the benchmarking trends reveal that Chinese LLMs are increasingly demonstrating strong performance, particularly in areas relevant to the Chinese market, such as Chinese language processing. Comparisons often focus on benchmarks assessing capabilities in mathematics, coding, reasoning, and multimodal understanding (4). The emergence of Chinese-specific benchmarks, such as SmartBench designed for evaluating on-device LLMs in real-world mobile scenarios, further underscores the maturing of the Chinese AI ecosystem (92).

Table 2: Select Benchmark Comparisons of Chinese LLMs (Early 2025)
Model MMLU (Pass@1) Codeforces (Percentile) MATH-500 (Pass@1) ArenaHard (vs GPT-4-1106)
DeepSeek R1 90.8 96.3 97.3 92.3% Win Rate
DeepSeek V3 88.5 58.7 90.2 85.5% Win Rate
OpenAI GPT-4o 88.7 23.6 76.6 83.8% Win Rate
Claude 3.5 Sonnet 88.3 20.3 78.3 80.7% Win Rate
Kimi k1.5 (Long-CoT) Not Specified 94 96.2 Not Specified
Kimi k1.5 (Short-CoT) Not Specified Not Specified 94.6 Not Specified
OpenAI o1 Not Specified Not Specified 94.8 Not Specified
Note: This table presents a selection of reported benchmark scores and is not exhaustive. Different benchmarks evaluate different aspects of model performance. Data primarily sourced from vendor announcements and related reports (e.g., 10, 14, 17, 28, 44, 47).

The rapid advancement of LLMs in China is occurring within a specific regulatory and ethical context. China has been among the first nations to actively regulate generative AI technologies, issuing the Generative AI Measures in July 2023 (93, 96). This framework establishes a multi-level approach encompassing laws and regulations related to data compliance, algorithm compliance, cybersecurity, and ethical considerations (95). Key regulatory requirements include content screening to ensure adherence to socialist values, the protection of personal data and privacy, the safeguarding of intellectual property rights, and the prevention of discriminatory practices in AI model design and training (93). Furthermore, online services utilizing algorithms that can influence public opinion are subject to algorithm filing and security assessment requirements (1). A central theme in China's regulatory approach is the alignment of AI development with national security and social stability objectives (65). Looking ahead, there is anticipation for the potential release of a draft AI Law in 2025, which would likely further solidify and strengthen the regulatory framework governing AI in China (95).

Ethical considerations are also playing an increasingly important role in shaping the development of AI in China (98, 99). The government has emphasized the importance of ethical standards, urging companies to ensure that their AI systems respect human rights and avoid discriminatory outcomes (97). Guidelines have been established to promote principles such as fairness, justice, harmony, safety, transparency, and accountability in the creation and deployment of AI systems (98). Despite these efforts, concerns remain regarding data privacy, the potential for inherent biases in training data, and the need for greater transparency in how AI models function (6). A particularly sensitive issue is the reported language-dependent censorship observed in some Chinese AI models, which raises ethical questions about the reliability and trustworthiness of the information they generate (101). Additionally, the Chinese government's control over the topics that AI models are permitted to address, shaping their responses to align with political boundaries, is a significant consideration (4). This complex interplay between fostering innovation and maintaining strict control is a defining characteristic of China's approach to AI regulation and ethics (100, 102).

The rapid advancements in Chinese LLMs are expected to have a profound impact on various industries and the broader society. Experts view LLMs as a critical engine for China's scientific and technological progress, driving industrial transformation and significantly boosting productivity (71). The potential to enhance efficiency, productivity, and foster innovation across a wide range of sectors is substantial (71, 77). These advanced AI models are anticipated to revolutionize business operations and significantly improve the experiences of consumers (82). By enabling AI-driven applications across diverse industries, they are advancing the fields of natural language processing and understanding at an unprecedented pace (1). Goldman Sachs Research has projected a substantial positive impact on China's GDP growth in the coming years, directly attributable to the widespread adoption and integration of AI technologies (103).

However, the societal implications of this rapid AI development are multifaceted. While there is considerable optimism, particularly among Chinese students who generally believe that the benefits of AI will outweigh the potential harms (83), concerns exist regarding the global preparedness to handle the ethical and safety challenges posed by such powerful technologies, especially within the context of authoritarian governance (6). The potential impact on employment is also a significant consideration, necessitating careful management to mitigate job displacement resulting from increased automation (103, 104). The use of AI in sensitive areas like surveillance and the development of social credit systems have raised human rights concerns internationally. Notably, the Chinese government aims to play a key role in shaping international norms and standards related to AI, including the management of potential risks (83). The ongoing advancements in Chinese LLMs are not only transforming domestic industries and society but are also poised to significantly influence the global AI landscape and the international balance of technological power (1).

The rapid progress in Chinese LLMs is being further accelerated by a network of collaborations and strategic partnerships within the domestic AI ecosystem. Major industry players are actively forming alliances to integrate advanced LLM capabilities into various applications. For instance, BMW and Alibaba have deepened their strategic partnership in China to incorporate Alibaba's Qwen large language model into BMW's next-generation intelligent vehicles (107, 108). Hon Hai Research Institute has expressed its intention to collaborate with technology partners to broaden the applications of its newly launched FoxBrain LLM (32). China Mobile International has launched the "AI+ Go Global Ecosystem Alliance" with the aim of fostering collaborative opportunities to leverage AI technology for the overseas expansion of Chinese enterprises (109). Additionally, Alibaba Cloud is collaborating with companies like HONOR and Zeekr to drive AI innovation in edge computing devices (107). These industry partnerships highlight a concerted effort to translate advancements in LLMs into tangible products and services across diverse sectors.

Research and development collaborations are also crucial to the advancement of Chinese LLMs. These efforts often involve partnerships between major technology companies, innovative startups, and leading academic institutions (1). Hon Hai Research Institute received significant technology assistance and consultation from NVIDIA during the development of its FoxBrain model (34). Zhipu AI has strategically partnered with hardware companies such as Huawei and Qualcomm to optimize its models for various platforms (54). These collaborations leverage the diverse expertise and resources available within the Chinese ecosystem to push the boundaries of AI capabilities.

Furthermore, there is a growing recognition of the importance of international engagement in the field of AI. While primarily focused on domestic development, some Chinese AI companies are beginning to participate in global initiatives related to AI safety and governance. For example, Zhipu.ai was the initial Chinese signatory to the Seoul Commitments for AI safety (110). The China Academy for Information and Communications Technology (CAICT) also announced China's own AI Safety Commitments, which bear notable similarities to the Seoul Commitments (110). Additionally, both DeepSeek and Moonshot have released open-source reasoning models that directly compete with offerings from OpenAI, indicating a willingness to engage with the global AI community (110, 111). These initial steps suggest a growing awareness of the need for international collaboration and alignment as Chinese LLMs continue to advance (106, 112).

Expert analysis of the recent advancements in Chinese LLMs suggests a significant turning point in the global AI landscape. The success of DeepSeek, in particular, has been viewed by many as a pivotal moment, potentially ushering in a new era of AI development in China (85, 86). Some experts believe that China has reached its "DeepSeek moment," predicting that 2025 will be a breakout year for the widespread application and deployment of advanced AI models across the country (85). The rise of DeepSeek has even led to the assertion that "closed-source AI has no future," with a growing belief that open-source development will be the primary driver of future progress in the field (85).

China's unique advantage, according to some analysts, lies in its solid and extensive industrial base and unified domestic market, which enables the development of more reliable AI solutions that are better aligned with the needs of the real economy, especially in areas like autonomous driving (85). These rapid advancements are also seen as directly challenging US policies and fundamentally reshaping the dynamics of the global AI industry (105).

Looking towards the future, experts predict that 2025 will be a year of "explosive growth" for AI agents and AI reasoning capabilities in China (84). Chinese AI companies are expected to continue increasing their capital expenditures and investments in this critical area (103). There is even the possibility that AI adoption rates in China will closely mirror those of developed economies, indicating a rapid pace of integration (103). However, challenges remain. Access to advanced semiconductor chips and the potential impact of US export restrictions continue to be significant concerns for the sustained development of cutting-edge models (4, 113). Within the scientific community, there is also an ongoing debate among Chinese researchers about whether the current singular focus on large language models represents the most effective long-term path towards achieving artificial general intelligence (AGI) (114, 115). Furthermore, the widespread adoption of AI will necessitate careful management of potential job displacement due to automation (103).

Expert opinions on the competitive landscape suggest a significant narrowing of the AI gap between China and the United States. Former Google CEO Eric Schmidt acknowledged in early 2025 that China has made remarkable progress and has largely caught up in the field of AI (5). Some analysts believe that DeepSeek's breakthrough is a direct response to US efforts to limit China's AI development, highlighting the resilience and innovation within the Chinese tech sector (1). Kai-Fu Lee, a prominent AI expert, has even predicted a consolidation within the Chinese LLM market, with DeepSeek, Alibaba, and ByteDance emerging as the three dominant players (25). These expert perspectives underscore the dynamic and rapidly evolving nature of the competition in the global AI arena.

11. Conclusion

The advancements in Chinese large language AI models announced in 2024 and early 2025 represent a significant leap forward in the global AI landscape. The rapid pace of innovation is evident in the emergence of highly competitive models like DeepSeek-R1, Alibaba's Qwen series, Baidu's Ernie series, and Moonshot AI's Kimi k1.5. These models showcase impressive technical specifications, including the adoption of efficient architectures like Mixture-of-Experts, the utilization of massive training datasets, and the incorporation of key features such as extended context windows and multimodal capabilities. Their reported capabilities span a wide array of applications across critical sectors like finance, healthcare, education, and manufacturing, highlighting their transformative potential. Performance benchmarks increasingly indicate that Chinese LLMs are achieving parity with, and in some cases surpassing, leading global models developed in the West. This progress is occurring within a unique regulatory and ethical environment in China, which prioritizes national interests and social stability while also grappling with the ethical implications of advanced AI. The impact of these advancements on industries and society is expected to be profound, driving economic growth and reshaping various aspects of daily life. Strategic collaborations and partnerships within the Chinese AI ecosystem are further fueling this rapid development. Expert analysis suggests that these recent breakthroughs mark a pivotal moment for China in the global AI race, with the potential to shift the balance of technological power and influence the future trajectory of artificial intelligence worldwide.

References

  1. The Geopolitics. (n.d.). *China's AI Revolution: The Global Tech Power Shift*. Retrieved March 31, 2025, from https://thegeopolitics.com/chinas-ai-revolution-the-global-tech-power-shift/
  2. Asia Society. (n.d.). *Implications of China's AI Strategy: State Engineering, Domestic Challenges, and Global Competition*. Policy Institute. Retrieved March 31, 2025, from https://asiasociety.org/policy-institute/implications-chinas-ai-strategy-state-engineering-domestic-challenges-and-global-competition
  3. INSEAD Knowledge. (n.d.). *Deeper Than DeepSeek: China's AI Ascendancy*. Retrieved March 31, 2025, from https://knowledge.insead.edu/strategy/deeper-deepseek-chinas-ai-ascendancy
  4. AP News. (n.d.). *DeepSeek has rattled the AI industry. Here's a quick look at other Chinese AI models*. Retrieved March 31, 2025, from https://apnews.com/article/china-ai-models-usa-technology-92d10dc20e3110b2774a5bc8f976e8f9
  5. Haq, R. (2025, January). *Chinese Lab's AI LLM Performance Shocks Silicon Valley*. Haq's Musings. Retrieved March 31, 2025, from http://www.riazhaq.com/2025/01/chinese-labs-ai-llm-performance-shocks.html
  6. Harvard International Review. (n.d.). *The Artificial Intelligence Revolution in an Unprepared World: China, the International Stage, and the Future of AI*. Retrieved March 31, 2025, from https://hir.harvard.edu/artificial-intelligence-china-and-the-international-stage/
  7. The Daily Cardinal. (2025, March). *Deepseek introduces new technologies to the AI world*. Retrieved March 31, 2025, from https://www.dailycardinal.com/article/2025/03/deepseek-introduces-new-technologies-to-the-ai-world
  8. Index.dev. (n.d.). *Top 6 Chinese AI Models Like DeepSeek (LLMs) You Should Know*. Retrieved March 31, 2025, from https://www.index.dev/blog/chinese-ai-models-deepseek
  9. EM360Tech. (n.d.). *What is DeepSeek? The Chinese AI Lab Rivalling OpenAI*. Retrieved March 31, 2025, from https://em360tech.com/tech-article/what-is-deepseek
  10. Live Science. (n.d.). *Chinese researchers just built an open-source rival to ChatGPT in 2 months. Silicon Valley is freaked out*. Retrieved March 31, 2025, from https://www.livescience.com/technology/artificial-intelligence/china-releases-a-cheap-open-rival-to-chatgpt-thrilling-some-scientists-and-panicking-silicon-valley
  11. YouTube. (n.d.). *How China's New AI Model DeepSeek Is Threatening U.S. Dominance*. Retrieved March 31, 2025, from https://www.youtube.com/watch?v=WEBiebbeNCA
  12. Shakudo. (2025, March). *Top 9 Large Language Models as of March 2025*. Retrieved March 31, 2025, from https://www.shakudo.io/blog/top-9-large-language-models
  13. Business Standard. (n.d.). *What is DeepSeek: China's open-source AI research lab which rivals OpenAI | World News*. Retrieved March 31, 2025, from https://www.business-standard.com/world-news/deepseek-r1-chinese-ai-research-breakthrough-challenging-openai-explained-125012700327_1.html
  14. GitHub. (n.d.). *DeepSeek-R1*. DeepSeek AI. Retrieved March 31, 2025, from https://github.com/deepseek-ai/DeepSeek-R1
  15. Hugging Face. (n.d.). *deepseek-ai/DeepSeek-R1*. Retrieved March 31, 2025, from https://huggingface.co/deepseek-ai/DeepSeek-R1
  16. Tech Wire Asia. (2024, September). *Chinese AI firms target growth as OpenAI introduces new LLM*. Retrieved March 31, 2025, from https://techwireasia.com/2024/09/chinese-ai-firms-target-growth-as-openai-introduces-new-llm/
  17. SCMP. (n.d.). *China's Alibaba releases new AI model, said to outperform competitors Deepseek and OpenAI's GPT-4o* [Video]. Retrieved March 31, 2025, from https://www.scmp.com/video/technology/3296789/chinas-alibaba-releases-new-ai-model-said-outperform-competitors-deepseek-and-openais-gpt-40
  18. Reddit. (n.d.). *ByteDance announces Doubao-1.5-pro*. r/LocalLLaMA. Retrieved March 31, 2025, from https://www.reddit.com/r/LocalLLaMA/comments/1i9wbya/bytedance_announces_doubao15pro/
  19. TMTPost. (n.d.). *ByteDance Valuation Said to Top $400 Billion, Driven by AI Business Doubao*. Retrieved March 31, 2025, from https://en.tmtpost.com/post/7471894
  20. Daily.dev. (n.d.). *ByteDance Introduces Doubao-1.5 Pro: A Cost-Effective and High-Performance AI Model*. Retrieved March 31, 2025, from https://app.daily.dev/posts/bytedance-introduces-doubao-1-5-pro-a-cost-effective-and-high-performance-ai-model-u6lbxhsk7
  21. Tech in Asia. (n.d.). *ByteDance launches AI model Doubao 1.5 Pro*. Retrieved March 31, 2025, from https://www.techinasia.com/news/bytedance-launches-ai-model-doubao-15-pro
  22. YouTube. (n.d.). *ByteDance's Doubao 1.5 Pro: The AI Game-Changer!* Retrieved March 31, 2025, from https://www.youtube.com/watch?v=4E7pR1gl7go
  23. Ashley. (n.d.). *ByteDance Just Dropped Doubao-1.5 Pro: The AI Model That's Quietly Outsmarting Everyone*. Towards AGI | Medium. Retrieved March 31, 2025, from https://medium.com/towards-agi/bytedance-just-dropped-doubao-1-5-pro-the-ai-model-thats-quietly-outsmarting-everyone-b9ebc523558d
  24. Analytics Drift. (n.d.). *ByteDance Launches an Advanced AI Model, Doubao-1.5-pro*. Retrieved March 31, 2025, from https://analyticsdrift.com/bytedance-launches-an-advanced-ai-model-doubao-1-5-pro/
  25. AIbase. (n.d.). *ByteDance Releases Doubao Large Model 1.5 Pro, Performance Surpassing GPT-4o and Claude3.5Sonnet*. Retrieved March 31, 2025, from https://www.aibase.com/news/14931
  26. Nerdynav. (2024, January). *Baidu Ernie 4.0 Vs ChatGPT: Abilities, Limits, Users*. Retrieved March 31, 2025, from https://nerdynav.com/baidu-ernie/
  27. AI Mode. (n.d.). *Ernie Bot AI: (What It Is, Features, Usecases, Price, Alternatives, How To Access)*. Retrieved March 31, 2025, from https://aimode.co/app/ernie-bot/
  28. Analytics Vidhya. (2025, March). *Baidu Releases ERNIE 4.5 & X1 Models Outperforming GPT-4.5 for 1% the Cost*. Retrieved March 31, 2025, from https://www.analyticsvidhya.com/blog/2025/03/ernie-4-5-x1/
  29. FastBots.ai. (n.d.). *What Is Ernie Bot: Introducing the Latest AI Technology*. Retrieved March 31, 2025, from https://fastbots.ai/blog/what-is-ernie-bot-introducing-the-latest-ai-technology
  30. Cybernews. (n.d.). *Baidu releases its first reasoning model ErnieX1 as AI competition ramps up*. Retrieved March 31, 2025, from https://cybernews.com/ai-news/baidu-releases-newernie-models/
  31. TechHQ. (2025, March). *China and the world's AI race: Baidu's new reasoning model*. Retrieved March 31, 2025, from https://techhq.com/2025/03/china-and-the-worlds-ai-race-baidus-new-reasoning-model/
  32. Hon Hai Precision Industry Co., Ltd. (Foxconn). (n.d.). *Hon Hai Research Institute Launches Traditional Chinese LLM With Reasoning Capabilities*. Retrieved March 31, 2025, from https://www.honhai.com/en-us/press-center/press-releases/latest-news/1549
  33. Canada Newswire. (n.d.). *Hon Hai Research Institute Launches Traditional Chinese LLM With Reasoning Capabilities*. Retrieved March 31, 2025, from https://www.newswire.ca/news-releases/hon-hai-research-institute-launches-traditional-chinese-llm-with-reasoning-capabilities-811143329.html
  34. Taipei Times. (n.d.). *Hon Hai launches its first advanced LLM 'FoxBrain'*. Retrieved March 31, 2025, from https://www.taipeitimes.com/News/biz/archives/2025/03/11/2003833206
  35. FT.com Markets data. (n.d.). *Hon Hai Research Institute Launches Traditional Chinese LLM With Reasoning Capabilities – Company Announcement*. Retrieved March 31, 2025, from https://markets.ft.com/data/announce/full?dockey=600-202503100616PR_NEWS_EURO_ND_EN37119-1
  36. Stock Titan. (n.d.). *Hon Hai Research Institute Launches Traditional Chinese LLM With Reasoning Capabilities*. Retrieved March 31, 2025, from https://www.stocktitan.net/news/HNHPF/hon-hai-research-institute-launches-traditional-chinese-llm-with-9tjhv4xfpkmz.html
  37. PRESS Insider. (n.d.). *Foxconn rolls out own AI model FoxBrain*. Retrieved March 31, 2025, from https://pressinsider.com/technology/foxconn-rolls-out-own-ai-model-foxbrain/
  38. R&D World. (n.d.). *Foxconn launches FoxBrain: AI for manufacturing in Traditional Chinese*. Retrieved March 31, 2025, from https://www.rdworldonline.com/foxconn-launches-foxbrain-ai-for-manufacturing-in-traditional-chinese/
  39. AI Mode. (n.d.). *Kimi K1.5*. Retrieved March 31, 2025, from https://aimode.co/model/kimi-k1-5/
  40. GitHub. (n.d.). *MoonshotAI/Kimi-k1.5*. Moonshot AI. Retrieved March 31, 2025, from https://github.com/MoonshotAI/Kimi-k1.5
  41. Reddy, N. (n.d.). *Understanding Kimi k1.5: Scaling Reinforcement Learning with LLMs*. Medium. Retrieved March 31, 2025, from https://medium.com/@nandinilreddy/understanding-kimi-k1-5-scaling-reinforcement-learning-with-llms-fc3294e0d943
  42. Appy Pie. (n.d.). *Kimi k1.5 vs DeepSeek R1: Battle of the Best Chinese LLMs*. Retrieved March 31, 2025, from https://www.appypie.com/blog/kimi-k15-vs-deepseek-r1
  43. The AI Track. (n.d.). *After DeepSeek-R1 Chinese Moonshot AI Launches Kimi k1.5*. Retrieved March 31, 2025, from https://theaitrack.com/moonshot-ai-launches-kimi-k1-5/
  44. Analytics Vidhya. (2025, January). *After DeepSeek, Kimi k1.5 Outshines OpenAI o1*. Retrieved March 31, 2025, from https://www.analyticsvidhya.com/blog/2025/01/kimi-k1-5/
  45. AI-Pro.org. (n.d.). *Kimi AI's Kimi k1.5 is Set to Elevate Multimodal AI*. Retrieved March 31, 2025, from https://ai-pro.org/learn-ai/articles/kimi-ai-elevates-multimodal-ai-with-kimi-k1-5/
  46. Greyling, C. (n.d.). *Kimi K1.5 Language Model*. Medium. Retrieved March 31, 2025, from https://cobusgreyling.medium.com/kimi-k1-5-language-model-5c420eac5766
  47. Reddit. (n.d.). *The Chinese AI model Kimi k1.5 has mastered reasoning and surpassed OpenAI o1*. r/Popular_Science_Ru. Retrieved March 31, 2025, from https://www.reddit.com/r/Popular_Science_Ru/comments/1if47rt/%D0%BA%D0%B8%D1%82%D0%B0%D0%B9%D1%81%D0%BA%D0%B0%D1%8F_%D0%B8%D0%B8%D0%BC%D0%BE%D0%B4%D0%B5%D0%BB%D1%8C_kimi_k15_%D0%BE%D1%81%D0%B2%D0%BE%D0%B8%D0%BB%D0%B0_%D1%80%D0%B0%D1%81%D1%81%D1%83%D0%B6%D0%B4%D0%B5%D0%BD%D0%B8%D1%8F_%D0%B8/?tl=en
  48. TechNode. (2025, January 9). *Meet China's top six AI unicorns: who are leading the wave of AI in China*. Retrieved March 31, 2025, from https://technode.com/2025/01/09/meet-chinas-top-six-ai-unicorns-who-are-leading-the-wave-of-ai-in-china/
  49. Topbots. (n.d.). *Beyond DeepSeek: An Overview of Chinese AI Tigers and Their Cutting-Edge Innovations*. Retrieved March 31, 2025, from https://www.topbots.com/chinese-ai-tigers-overview/
  50. Madan, P. (n.d.). *This AI Model is paving way for Multilingual enterprises use cases: GLM-4 Plus*. Medium. Retrieved March 31, 2025, from https://medium.com/@parasmadan.in/this-ai-model-is-paving-way-for-multilingual-enterprises-use-cases-glm-4-plus-916834d40614
  51. ZHIPU AI OPEN PLATFORM. (n.d.). *GLM-4-Plus*. Retrieved March 31, 2025, from https://bigmodel.cn/dev/howuse/glm-4
  52. ZHIPU AI OPEN PLATFORM. (n.d.). *Overview*. Retrieved March 31, 2025, from https://bigmodel.cn/dev/howuse/model
  53. ZHIPU AI OPEN PLATFORM. (n.d.). *ZHIPU AI OPEN PLATFORM*. Retrieved March 31, 2025, from https://open.bigmodel.cn/
  54. Data Vidersity. (2024, December). *Zhipu AI: China's Generative Trailblazer Grappling with Rising Competition*. Retrieved March 31, 2025, from https://datainnovation.org/2024/12/zhipu-ai-chinas-generative-trailblazer-grappling-with-rising-competition/
  55. The Economic Times. (n.d.). *China's Zhipu AI launches free AI agent, intensifying domestic tech race*. Retrieved March 31, 2025, from https://m.economictimes.com/tech/artificial-intelligence/chinas-zhipu-ai-launches-free-ai-agent-intensifying-domestic-tech-race/articleshow/119808131.cms
  56. ZHIPU AI OPEN PLATFORM. (n.d.). *Product Pricing*. Retrieved March 31, 2025, from https://open.bigmodel.cn/pricing
  57. Reddit. (n.d.). *GLM-4-Voice: Zhipu AI's New Open-Source End-to-End Speech Large Language Model*. r/LocalLLaMA. Retrieved March 31, 2025, from https://www.reddit.com/r/LocalLLaMA/comments/1gbzbnp/glm4voice_zhipu_ais_new_opensource_endtoend/
  58. Epoch AI. (n.d.). *What went into training DeepSeek-R1?* Gradient Updates. Retrieved March 31, 2025, from https://epoch.ai/gradient-updates/what-went-into-training-deepseek-r1
  59. GitHub. (n.d.). *FareedKhan-dev/train-deepseek-r1*. Retrieved March 31, 2025, from https://github.com/FareedKhan-dev/train-deepseek-r1
  60. Kili Technology. (n.d.). *Qwen 2: What can we learn about Alibaba's Multilingual LLM?* Retrieved March 31, 2025, from https://kili-technology.com/large-language-models-llms/qwen-2-guide-what-can-we-learn-about-alibaba-s-multilingual-llm
  61. Inferless. (n.d.). *Qwen Models: Alibaba's Next-Generation AI Family for Text, Vision, and Beyond*. Retrieved March 31, 2025, from https://www.inferless.com/learn/the-ultimate-guide-to-qwen-model
  62. Qwen Team. (n.d.). *Qwen*. Read the Docs. Retrieved March 31, 2025, from https://qwen.readthedocs.io/
  63. AI Mode. (n.d.). *Baidu Ernie Bot Statistics: Latest Trends And Performance Analysis*. Retrieved March 31, 2025, from https://aimode.co/baidu-ernie-bot-statistics/
  64. DataCamp. (n.d.). *Qwen (Alibaba Cloud) Tutorial: Introduction and Fine-Tuning*. Retrieved March 31, 2025, from https://www.datacamp.com/tutorial/qwen-alibaba-cloud
  65. Credence Research. (n.d.). *China Large Language Model Market Size, Growth & Forecast 2032*. Retrieved March 31, 2025, from https://www.credenceresearch.com/report/china-large-language-model-market
  66. EurekAlert!. (n.d.). *RA-CFGPT: Chinese financial assistant with retrieval-augmented large language model*. Retrieved March 31, 2025, from https://www.eurekalert.org/news-releases/1068029
  67. Yicai Global. (n.d.). *China's Financial Sector Hastens Integration of DeepSeek LLMs*. Retrieved March 31, 2025, from https://www.yicaiglobal.com/news/chinas-financial-sector-hastens-integration-of-deepseek-llms
  68. arXiv. (n.d.). *SuperCLUE-Fin: Graded Fine-Grained Analysis of Chinese LLMs on Diverse Financial Tasks and Applications*. Retrieved March 31, 2025, from https://arxiv.org/html/2404.19063v1
  69. Portfolio Management Research. (n.d.). *Large Language Models for Financial and Investment Management: Applications and Benchmarks*. The Journal of Portfolio Management. Retrieved March 31, 2025, from https://www.pm-research.com/content/iijpormgmt/51/2/162
  70. Portfolio Management Research. (n.d.). *Large Language Models for Financial and Investment Management: Models, Opportunities, and Challenges*. The Journal of Portfolio Management. Retrieved March 31, 2025, from https://www.pm-research.com/content/iijpormgmt/51/2/211
  71. Global Times. (n.d.). *China's LLMs gearing up to boost industrial development*. Retrieved March 31, 2025, from https://www.globaltimes.cn/page/202407/1315511.shtml
  72. Journal of Medical Internet Research. (2025). *Revolutionizing Health Care: The Transformative Impact of Large Language Models in Medicine*. *JMIR*, *1*(1), e59069. https://www.jmir.org/2025/1/e59069
  73. Sage Journals. (2025). Application of large language models in healthcare: A bibliometric analysis. *Digital Health*. https://doi.org/10.1177/20552076251324444
  74. PubMed. (n.d.). *Application of large language models in healthcare: A bibliometric analysis*. Retrieved March 31, 2025, from https://pubmed.ncbi.nlm.nih.gov/40035041/?utm_source=WordPress&utm_medium=rss&utm_campaign=pubmed-2&utm_content=1hUStbo8LeOX_JB_iJwSZLVmHUltDGhdWVlhWBkeLPeb19a9a8&fc=20221207123816&ff=20250304114743&v=2.18.0.post9+e462414
  75. arXiv. (n.d.). *MedGo: A Chinese Medical Large Language Model*. Retrieved March 31, 2025, from https://arxiv.org/html/2410.20428v1
  76. ResearchGate. (n.d.). *(PDF) Large language models in medical and healthcare fields: applications, advances, and challenges*. Retrieved March 31, 2025, from https://www.researchgate.net/publication/384161687_Large_language_models_in_medical_and_healthcare_fields_applications_advances_and_challenges
  77. VOA News. (n.d.). *China boosting development of AI for use in trade war with US*. Retrieved March 31, 2025, from https://www.voanews.com/a/china-boosting-development-of-ai-for-use-in-trade-war-with-us/8007739.html
  78. Birkins, J. (n.d.). *Generative AI and LLM in Education: 25 Use Cases, 14 applications*. Medium. Retrieved March 31, 2025, from https://medium.com/@joycebirkins/generative-ai-and-llm-in-education-25-use-cases-14-applications-a93ff73260cf
  79. arXiv. (n.d.). *Towards Evaluating the Chinese Education Values of Large Language Models*. Retrieved March 31, 2025, from https://arxiv.org/html/2409.12739v3
  80. PubMed Central (PMC). (n.d.). *Exploring the potential of LLM to enhance teaching plans through teaching simulation*. Retrieved March 31, 2025, from https://pubmed.ncbi.nlm.nih.gov/39910074/
  81. arXiv. (n.d.). *An Empirical Study on Large Language Models in Accuracy and Robustness under Chinese Industrial Scenarios*. Retrieved March 31, 2025, from https://arxiv.org/html/2402.01723v1
  82. People's Daily Online. (2024, December 31). *China's AI tech charts courses beyond ChatGPT*. Retrieved March 31, 2025, from http://en.people.cn/n3/2024/1231/c90000-20260756.html
  83. Jamestown Foundation. (n.d.). *Survey: How Do Elite Chinese Students Feel About the Risks of AI?* Retrieved March 31, 2025, from https://jamestown.org/program/survey-how-do-elite-chinese-students-feel-about-the-risks-of-ai/
  84. Global Times. (n.d.). *Chinese start-up launches new AI agent amid sector's 'explosive growth'*. Retrieved March 31, 2025, from https://www.globaltimes.cn/page/202503/1331253.shtml
  85. Chinadaily.com.cn. (2025, March 31). *Experts praise country's vision of AI development*. Retrieved March 31, 2025, from https://www.chinadaily.com.cn/a/202503/31/WS67e96ed7a3101d4e4dc2bb10.html
  86. People's Daily Online. (2025, March 31). *Experts praise country's vision of AI development*. Retrieved March 31, 2025, from http://en.people.cn/n3/2025/0331/c90000-20295855.html
  87. Analytics India Magazine. (n.d.). *Chinese Startup Launches Free AI Agent Powered by Models 8x Faster than DeepSeek-R1*. Retrieved March 31, 2025, from https://analyticsindiamag.com/ai-news-updates/chinese-startup-launches-free-ai-agent-powered-by-models-8x-faster-than-deepseek-r1/
  88. Caixin Global. (2025, March 31). *Chinese Startup Zhipu Takes Aim at OpenAI With Free AI Agent*. Retrieved March 31, 2025, from https://www.caixinglobal.com/2025-03-31/chinese-startup-zhipu-takes-aim-at-openai-with-free-ai-agent-102304449.html
  89. StartupNews.fyi. (2025, April 1). *China's Zhipu AI launches free AI agent, intensifying domestic tech race*. Retrieved March 31, 2025, from https://startupnews.fyi/2025/04/01/chinas-zhipu-ai-launches-free-ai-agent-intensifying-domestic-tech-race-2/
  90. Originality.ai. (n.d.). *Baidu Ernie Bot Statistics*. Retrieved March 31, 2025, from https://originality.ai/blog/baidu-ernie-bot-statistics
  91. KI Company. (n.d.). *The 5 best open-source AI models 2025 — DeepSeek leads*. Retrieved March 31, 2025, from https://www.ki-company.ai/en/blog-beitraege/the-5-best-open-source-ai-models-in-2025
  92. arXiv. (n.d.). *SmartBench: Is Your LLM Truly a Good Chinese Smartphone Assistant?* Retrieved March 31, 2025, from https://arxiv.org/html/2503.06029v1
  93. Reed Smith LLP. (2024, October). *AI explained: AI regulations and PRC court decisions in China*. Retrieved March 31, 2025, from https://www.reedsmith.com/en/perspectives/2024/10/ai-explained-ai-regulations-and-prc-court-decisions-in-china
  94. Cambridge University Press. (n.d.). Navigating China's regulatory approach to generative artificial intelligence and large language models. *Cambridge Forum on AI: Law and Governance*. Retrieved March 31, 2025, from https://www.cambridge.org/core/journals/cambridge-forum-on-ai-law-and-governance/article/navigating-chinas-regulatory-approach-to-generative-artificial-intelligence-and-large-language-models/969B2055997BF42DE693B7A1A1B4E8BA
  95. Law.asia. (n.d.). *Charting a course with AI regulation*. Retrieved March 31, 2025, from https://law.asia/ai-regulation-asia-generative-models-misuse/
  96. Twobirds.com. (2025). *China TMT: Annual Review of 2024 and Outlook for 2025 (I)*. Retrieved March 31, 2025, from https://www.twobirds.com/en/insights/2025/china/china-tmt-annual-review-of-2024-and-outlook-for-2025-(i)
  97. CBBC Focus. (n.d.). *What are China's new guidelines on AI regulation?* Retrieved March 31, 2025, from https://focus.cbbc.org/what-are-chinas-new-guidelines-on-artificial-intelligence-regulation/
  98. China Law Vision. (2025, January). *AI Ethics: Overview (China)*. Retrieved March 31, 2025, from https://www.chinalawvision.com/2025/01/digital-economy-ai/ai-ethics-overview-china/
  99. Cao, J., & Meng, T. (2025). Ethics and governance of artificial intelligence in digital China: Evidence from online survey and social media data. *Sage Journals*. https://doi.org/10.1177/2057150X241313085
  100. Fintech Weekly. (n.d.). *Ethical Considerations in the Deployment of DeepSeek AI in Fintech*. Retrieved March 31, 2025, from https://www.fintechweekly.com/magazine/articles/deepseek-in-fintech-ethical-considerations
  101. AInvest. (n.d.). *AI's Answers on China Vary by Language, Raising Ethical Questions*. Retrieved March 31, 2025, from https://www.ainvest.com/news/ai-answers-china-vary-language-raising-ethical-questions-2503
  102. CEPA. (n.d.). *China Walks Perilous AI Tightrope*. Retrieved March 31, 2025, from https://cepa.org/article/china-walks-perilous-ai-tightrope/
  103. Goldman Sachs. (n.d.). *What advanced AI means for China's economic outlook*. Insights. Retrieved March 31, 2025, from https://www.goldmansachs.com/insights/articles/what-advanced-ai-means-for-chinas-economic-outlook
  104. Macau Business. (n.d.). *China's rapid AI growth sparks hiring boom as demand outpaces supply*. Retrieved March 31, 2025, from https://www.macaubusiness.com/chinas-rapid-ai-growth-sparks-hiring-boom-as-demand-outpaces-supply/
  105. The George Washington University. (n.d.). *Media Tip Sheet: China's AI Breakthroughs Challenge U.S. Policies and Reshape Global Industry Dynamics*. GW Media Relations. Retrieved March 31, 2025, from https://mediarelations.gwu.edu/media-tip-sheet-chinas-ai-breakthroughs-challenge-us-policies-and-reshape-global-industry-dynamics
  106. The AI Track. (n.d.). *China's Rising Dominance in Large-Language Models (LLM)*. Retrieved March 31, 2025, from https://theaitrack.com/hina-dominance-large-language-models/
  107. Alibaba Cloud. (n.d.). *BMW and Alibaba Deepen Strategic Partnership in China, Harnessing Qwen's AI Power to Redefine Intelligent In-Car Experiences*. Blog. Retrieved March 31, 2025, from https://www.alibabacloud.com/blog/bmw-and-alibaba-deepen-strategic-partnership-in-china-harnessing-qwens-ai-power-to-redefine-intelligent-in-car-experiences_602094
  108. Xinhua. (n.d.). *BMW, Alibaba team up to integrate AI into next-generation vehicles*. Retrieved March 31, 2025, from https://english.news.cn/20250326/7f916bde1a6f4de285590399d4ad030c/c.html
  109. Laotian Times. (2025, March 20). *China Mobile International Launched "AI+ Global Solutions" and "CMI AI LLM Integrated Server" at AI+ Era Global Development Forum*. Retrieved March 31, 2025, from https://laotiantimes.com/2025/03/20/china-mobile-international-launched-ai-global-solutions-and-cmi-ai-llm-integrated-server-at-ai-era-global-development-forum/
  110. Carnegie Endowment for International Peace. (2025, January). *DeepSeek and Other Chinese Firms Converge with Western Companies on AI Promises*. Retrieved March 31, 2025, from https://carnegieendowment.org/research/2025/01/deepseek-and-other-chinese-firms-converge-with-western-companies-on-ai-promises?lang=en
  111. Atlantic Council. (n.d.). *What DeepSeek's breakthrough says (and doesn't say) about the 'AI race' with China*. New Atlanticist Blog. Retrieved March 31, 2025, from https://www.atlanticcouncil.org/blogs/new-atlanticist/what-deepseeks-breakthrough-says-and-doesnt-say-about-the-ai-race-with-china/
  112. MERICS. (n.d.). *Large language model development in China thrives, but geopolitics may spell trouble*. Comment. Retrieved March 31, 2025, from https://merics.org/en/comment/large-language-model-development-china-thrives-geopolitics-may-spell-trouble
  113. chinaobservers.eu. (n.d.). *Chinese LLM Developers Explore Domestic Chips Options*. Retrieved March 31, 2025, from https://chinaobservers.eu/chinese-llm-developers-explore-domestic-chips-options/
  114. Center for Security and Emerging Technology (CSET). (n.d.). *Chinese Critiques of Large Language Models*. Georgetown University. Retrieved March 31, 2025, from https://cset.georgetown.edu/publication/chinese-critiques-of-large-language-models/
  115. Center for Security and Emerging Technology (CSET). (n.d.). *Chinese Critiques of Large Language Models: Finding the Path to General Artificial Intelligence* [PDF]. Georgetown University. Retrieved March 31, 2025, from https://cset.georgetown.edu/wp-content/uploads/CSET-Chinese-Critiques-of-Large-Language-Models-Finding-the-Path-to-General-Artificial-Intelligence.pdf

Comments

Post a Comment

Sign Up For Our Free Newsletter & Vip List