阿里Qwen3.8正面硬刚Kimi K3：参数战打到2.4万亿 谁慢谁出局

This article summarizes the latest key updates on China's large language model (LLM) competition. Key takeaways for general users are as follows:

1. On August 3, Alibaba officially launched Qwen 3.8, the most powerful model in its Qwen family. With a total of 2.4 trillion parameters, it ranks among the world's top-tier LLMs, placing second globally on multi-modal performance rankings. Alibaba also launched Qianwen Office, an Agent-based productivity product, which will soon add a direct entry in DingTalk, allowing general users to access it at low cost in the near future.

2. The model significantly improves efficiency for daily work and content processing. It can complete what takes a human a week in just one hour — for example, annotating hundreds of legal documents. It can annotate 160 hours of basketball game footage in dozens of minutes, parse 200-page financial reports in one go, and organize content from 100-hour long videos, saving users a great deal of time on repetitive work.

3. Qwen 3.8 is priced accessibly: in the Chinese market, it costs 12 yuan per million input tokens and 36 yuan per million output tokens, with only 1.5 yuan per million tokens for cache hits. Its Max version will be open-sourced next week, allowing AI enthusiasts to download and test it for free.

This article outlines the current competitive landscape and development trends of China's LLM industry, providing actionable insights for brands looking to integrate AI into their operations. Key takeaways are as follows:

1. LLM competition has entered a phase defined by both parameter scaling and real-world落地 capability. Top-tier Chinese LLMs now can independently deliver long-term, complex projects, and can empower multiple business links for brands — including legal clause review, marketing material processing, and user behavior data analysis — delivering substantial cost reduction and efficiency gains.

2. On the application side, LLMs have shifted from general-purpose question answering to落地 in office and professional scenarios. Brands can access leading models like Qwen to optimize internal workflows, boost content production efficiency, improve customer service experience, and cut labor costs.

3. Calling costs for Chinese LLMs have dropped dramatically: international pricing is only 20% to 40% of that charged by leading overseas models, and domestic pricing is also very accessible. This has drastically lowered the barrier to AI adoption for brands, even allowing small and medium-sized brands to test integration.

This article discloses the latest competitive landscape and落地 progress of China's LLM industry, offering valuable insights for sellers looking to leverage AI and capture new opportunities. Key takeaways are as follows:

1. Overall industry opportunity: Chinese LLM technology has reached maturity. As of July, all top six most frequently called LLMs on global rankings are Chinese open-source models, with performance matching the global top tier, and low access costs. Sellers can use LLMs to optimize end-to-end operations including product selection, product copywriting generation, customer service, and market data analysis, boosting operational efficiency.

2. Latest accessible resources: Alibaba's Qwen 3.8 is already available via API, and is integrated into the DingTalk ecosystem. Qianwen Office is already in open beta, with a DingTalk entry launching soon, allowing sellers to access it at low cost. Sellers with in-house technical capabilities can also deploy the model locally to adapt to their own business after it is open-sourced next week.

3. Risk warning: The LLM industry is seeing extremely rapid technological iteration, with the competitive landscape shifting fundamentally every 1 to 2 months. Sellers should choose mature, leading models for their AI initiatives, to avoid business disruptions from outdated technology.

This article discloses the latest capability upgrades of leading LLMs, providing reference for factories advancing digital transformation and exploring new business opportunities. Key takeaways are as follows:

1. LLMs can now empower product manufacturing and R&D for factories. Leading LLMs are already capable of independently completing complex professional tasks: Kimi K3 can finish the entire chip design process without human intervention, while Alibaba's Qwen 3.8 can independently complete programming projects and streamline chip hardware gate counts. Factories can leverage LLMs to shorten R&D cycles and cut R&D labor costs.

2. Insights for digital and e-commerce transformation: The cost of LLM落地 has dropped significantly. Qwen 3.8 is priced accessibly, and its core model will be open-sourced. Factories do not need to invest heavily in custom development to build their own digital R&D and production management systems tailored to their business needs.

3. New business opportunities: China's LLM ecosystem is maturing rapidly. Factories can partner with LLM providers to develop vertical applications tailored to their own manufacturing scenarios, unlock new growth, and build differentiated competitive advantages.

This article summarizes the latest industry trends in the LLM space, offering high reference value for AI service providers to track industry shifts and develop customer solutions. Key takeaways are as follows:

1. Industry development trends: LLM competition has shifted from pure parameter competition and benchmark leaderboarding to competition over real-world落地 capability. The core battlefield has moved from general question answering to programming, long-horizon tasks, and professional office scenarios. Parameter competition has also made a comeback, with all leading players now joining the "2-trillion-parameter club", and technological innovation is accelerating. The competitive landscape restructures every 1 to 2 months, so service providers need to keep their solutions updated aligned with cutting-edge advances from leading players.

2. Core customer pain points: Customers are no longer satisfied with basic question answering capabilities; they require LLMs that can independently complete complex long-term projects and boost productivity for real-world business operations. For example, enterprises need to process batches of legal documents, office materials and R&D tasks, so demand for low-cost, high-performance solutions is strong.

3. Ready-to-use integration opportunities: Alibaba's Qwen 3.8 is already available via API, will be open-sourced next week, and ships with a multi-modal Agent extension library that is compatible with Alibaba's high-performance computing infrastructure. Service providers can build vertical scenario solutions quickly based on this model to cut their own development costs.

This article discloses the latest industry developments in the LLM space, offering reference for operation and strategic planning of LLM platforms and various industrial platforms. Key takeaways are as follows:

1. Current market demand from platforms: Both developers and enterprise users need high-performance, low-cost open-source LLMs. Leading models now can independently complete long-term complex tasks, and demand for open, cost-effective top-tier models is strong. Platforms need to add these models in a timely manner to meet user demand and improve their own competitiveness.

2. Referenceable best practice: Alibaba launched the Qianwen Office Agent product alongside its new LLM release, integrated a direct entry into DingTalk, and aligned model capability upgrades with product落地 to turn technological breakthroughs into tangible productivity users can experience. This model of pairing technological advancement with scenario development is worth learning for platforms.

3. Risks to avoid: The LLM industry is seeing extremely fast technological iteration, with product update cycles of only 1 to 2 months. Platforms need to update and add leading models in a timely manner to avoid being abandoned by users for outdated technology. They also need to prioritize cost-effectiveness, as competitive pricing attracts more users and API calls, boosting platform activity.

This article documents the latest competitive dynamics of China's LLM industry, providing valuable source material for AI industry researchers. Key findings are as follows:

1. Latest industry trends: Chinese LLMs have advanced rapidly and now rank among the global top tier. As of July, all top six most frequently called LLMs on OpenRouter's ranking are open-source models developed independently by Chinese firms. Competition has shifted focus from general question answering to programming and long-horizon Agent tasks. Parameter competition has returned, all leading players have entered the 2-trillion-parameter club, technological update cycles have shortened to 1 to 2 months, and competition has clearly accelerated.

2. New characteristics of industry development: LLM technology competition has moved beyond benchmark leaderboarding to competition over real project落地 capability. Leading models can now independently deliver long-term complex projects that take more than ten days to complete, marking a shift toward technology that transforms actual production methods.

3. New business model exploration: Leading LLM developers generally adopt an "open-source + paid API" model, pursuing a high cost-performance strategy. They also advance both C-end user access and enterprise级落地 simultaneously, integrate access through office instant messaging portals, and explore commercialization paths for technology落地, providing new business model samples for industry research.