Investment Rating - The report does not explicitly state an investment rating for the industry Core Insights - The competition in AI entry points is intensifying, with major companies increasing their investments. China's AI presence globally has significantly improved, with domestic large models continuously iterating. Despite GPT-5.2 and Gemini 3 Pro leading, Chinese models have effectively altered the North American dominance in the competitive landscape. In the global Top 10, three positions are held by Chinese models, and in the Top 15, there are six Chinese companies. By 2025, China's open-source AI model usage is expected to account for over 70% of the global market [2][11][19] - The demand for inference has surged, with the emergence of o1 class inference models unlocking approximately 10 times the potential of traditional models in terms of inference-time compute. The demand for computing power has shifted from being solely "training-driven" to a dual focus on "training + inference" [2][5][37] - The battle for entry points has evolved beyond mobile devices to OS-level intelligent agents and super apps. By December 24, 2025, ByteDance's AI application Doubao announced daily active users (DAU) exceeding 100 million, while Qianwen App reached over 30 million monthly active users within 23 days of public testing, becoming the fastest-growing AI application globally. Doubao bypasses traditional interfaces, creating an "AI operating system" that directly interacts with super apps like WeChat and Alipay, challenging the rules of the traditional app era [2][44][45] Summary by Sections AI Entry Point Competition - China's AI global presence has significantly improved, with domestic large models continuously iterating. In the global Top 10, three positions are held by Chinese models, and in the Top 15, there are six Chinese companies. By 2025, China's open-source AI model usage is expected to account for over 70% of the global market [2][11][19] - The competition for entry points has evolved beyond mobile devices to OS-level intelligent agents and super apps, with significant user engagement reported for new AI applications [2][44][45] Domestic Chip Breakthroughs - The smart computing center in China is expanding, with a projected compound annual growth rate (CAGR) of 57% from 2020 to 2028, reaching 2,781.9 EFLOPS by 2028. Domestic chip technology is steadily improving, with local cloud service providers accelerating the construction of heterogeneous environments [5][50] - Domestic general-purpose GPUs are upgrading from "usable" to "good," with performance metrics approaching those of leading international models. The production capacity of domestic chip manufacturers like SMIC is continuously increasing, providing solid support for domestic AI chip production [5][53][54] Supply and Demand Dynamics - The demand side is characterized by a surge in inference demand as AI applications become more prevalent, while the supply side sees continuous improvements in domestic GPU performance and accelerated adaptation by cloud service providers [5][59] - The AI server market is expected to see a shift towards inference servers becoming the mainstream, with a projected market size of approximately $39.3 billion in 2024, reflecting a year-on-year growth of 49.7% [5][64]

Core Insights - The article highlights the impressive performance of MiniMax's new model M2 in the mini-SWE-agent benchmark, surpassing competitors like DeepSeek, GLM, Qwen, and Kimi [2][4] - MiniMax M2's success is attributed to its innovative "Interleaved Thinking" approach, which allows for simultaneous reasoning and tool usage, enhancing its ability to handle complex tasks [4][5] Performance and Recognition - MiniMax M2 has received widespread recognition from developers within just over a month of its release, demonstrating its effectiveness in real-world agent applications [5] - The model's ability to maintain context and improve self-correction capabilities has been noted as a significant advantage, leading to better planning and execution in complex tasks [5][25] Interleaved Thinking Mechanism - Interleaved Thinking is a new reasoning paradigm that integrates reasoning and action, addressing limitations of traditional linear models [10][11] - This approach allows for a dynamic cycle of "thinking → acting → observing → rethinking," which significantly enhances the reliability of long-term workflows [12][25] - The technique effectively mitigates "state drift," ensuring that plans and intentions can persist across multiple interactions, which is crucial for complex agent tasks [16][17] Comparison with Other Memory Techniques - Interleaved Thinking differs from traditional memory models by focusing on maintaining logical reasoning rather than just factual recall, akin to a computer's RAM [20] - While traditional models store past interactions, Interleaved Thinking preserves the reasoning process, enabling agents to make informed decisions based on previous steps [21] Industry Adoption and Future Implications - The adoption of Interleaved Thinking is becoming a standard in high-performance agent models, with other leading companies also integrating similar capabilities [22][23] - MiniMax M2 is positioned as a pioneer in this technology, showcasing unique methods to enhance performance and efficiency [23][25] Cost Efficiency and Practical Applications - MiniMax M2 demonstrates remarkable cost efficiency, with a total operational cost of $0.001669 for a complex task, significantly lower than competitors [31] - This economic advantage allows developers to conduct more iterations within the same budget, facilitating rapid experimentation and development [31] Community and Ecosystem Development - MiniMax is actively working to standardize the implementation of Interleaved Thinking through collaborations with various partners and providing best practices for developers [38][39] - The introduction of tools like the Mini-Agent CLI aims to help developers effectively utilize Interleaved Thinking in their projects, enhancing community engagement and support [44][46]

Core Insights - DeepSeek has released two official models: DeepSeek-V3.2 and DeepSeek-V3.2-Speciale, with the former focusing on balancing reasoning ability and output length for everyday use, while the latter enhances long-form reasoning and mathematical proof capabilities [1][2][4] - The open-source large model ecosystem has seen significant growth, with DeepSeek's advancements posing a challenge to closed-source models, particularly in light of the recent release of Google Gemini 3.0, which has raised the competitive bar [2][15] - DeepSeek's models are positioned to bridge the gap between open-source and closed-source models through innovative architecture and training strategies, despite limitations in computational resources compared to industry giants [8][15][16] Model Performance - DeepSeek-V3.2 has achieved performance levels comparable to GPT-5 and is slightly below Google’s Gemini 3 Pro, demonstrating its effectiveness in reasoning tasks [6][7] - The Speciale version has outperformed Gemini 3 Pro in several reasoning benchmarks, including the American Mathematics Invitational Exam (AIME) and the Harvard-MIT Mathematics Tournament (HMMT) [7][8] - Speciale's design focuses on rigorous mathematical proof and logical verification, making it a specialized tool for complex reasoning tasks [6][8] Technological Innovations - DeepSeek employs a novel DSA (DeepSeek Sparse Attention) mechanism to optimize computational efficiency, allowing for effective long-context processing without sacrificing performance [8][12] - The concept of "Interleaved Thinking" has been integrated into DeepSeek's models, enhancing the interaction between reasoning and tool usage, which is crucial for AI agents [9][12] - The focus on agent capabilities signifies a strategic shift towards creating actionable AI, moving beyond traditional chat-based interactions to more complex task execution [13][14] Industry Context - The competitive landscape is shifting, with DeepSeek acknowledging the widening gap between open-source and closed-source models, particularly in complex task performance [15][16] - DeepSeek aims to address its limitations by increasing pre-training computational resources and optimizing model efficiency, indicating a clear path for future improvements [16][19] - The release of DeepSeek-V3.2 has been seen as a significant achievement in the open-source community, suggesting that the gap with leading closed-source models is narrowing [16][19]