腾讯Hy3周使用量涨68倍 拟推出更大参数Hy4模型

This article discloses the latest progress of Tencent's self-developed Hunyuan large model, with key takeaways as follows:

1. Tencent has launched the official version of Hunyuan Hy3, a general-purpose multimodal large model that supports text, images, video, 3D and other modalities. Built on a MoE architecture, it has a total of 295 billion parameters, with 21 billion activated parameters, and supports a 256K context window. Its performance matches that of leading domestic and international flagship models with 2 to 5 times its total parameter size. Since launch, its weekly usage has increased 68-fold compared to the previous generation.

2. General users can access Hy3 through multiple Tencent internal products, including the office tool WorkBuddy, coding tool CodeBuddy, and AI assistant Yuanbao. Users in overseas regions can enjoy free access to Hy3-related features until August 31, 2026. Developers worldwide can also commercially utilize the model via APIs and open-source licenses.

3. Tencent plans to launch Hy4, a larger model with stronger performance, though the specific release date and parameter scale have not yet been announced.

This article discloses the latest updates on Tencent's Hunyuan large model, providing valuable references for brands looking to leverage AI to empower their business. Key takeaways are as follows:

1. The fully iterated Hunyuan Hy3 is now officially launched and commercially available, with performance reaching industry flagship levels. It supports a wide range of brand operation scenarios including content creation, knowledge Q&A, and business automation. Brands can integrate the model's capabilities to optimize marketing content production, user consulting services, and internal operation automation, cutting costs while boosting efficiency.

2. Tencent is continuing to scale up its AI investments: in Q2 2026, its R&D spending increased 35% year-over-year, and capital expenditure rose 176% year-over-year, with the majority of investment allocated to AI computing power procurement and large model R&D. A more powerful Hy3 model is already in the pipeline, meaning technical capabilities and the supporting ecosystem will continue to improve.

3. AI empowerment has become a prevailing trend in the consumer and service sectors. Brands can leverage mature large model capabilities to quickly implement intelligent upgrades and capture the next wave of user growth.

This article shares the latest updates on the commercial deployment of Tencent's large model, bringing the following opportunities and insights for sellers:

1. Tencent's Hy3 large model is now officially launched, with commercial access via APIs and open-source licensing available to global developers. Sellers can integrate the model to build custom tools for intelligent customer service, marketing content generation, product selection analysis, and operation automation, effectively cutting operating costs and improving efficiency.

2. Overseas users can access Hy3 for free until the end of August 2026, giving cross-border sellers an opportunity to test the model's capabilities and its suitability for overseas user service scenarios at no upfront cost, helping them identify new growth opportunities for cross-border business and reduce early technical trial-and-error costs.

3. Tencent continues to ramp up AI investment and will launch the more powerful Hy4 large model in the future, with AI capabilities iterating at a rapid pace. Sellers can proactively integrate large model capabilities now to capitalize on AI-driven opportunities in retail e-commerce and build differentiated competitive advantages.

The industry updates on Tencent's large model outlined in this article offer the following insights for factories advancing digital and intelligent transformation:

1. General-purpose multimodal large models have now reached full maturity, supporting text, images, video, 3D and other modalities to power a wide range of enterprise scenarios from content creation to business automation. Factories can leverage these capabilities to optimize product design processes: for example, 3D modality support allows factories to complete product modeling and prototyping faster, cutting product design cycles and improving design efficiency.

2. Large models are now commercially available, and leading tech players have packaged their capabilities as standardized services. This has drastically lowered the barrier to access for factories, which can now adopt mature AI capabilities to advance digital transformation without bearing the high cost of developing a large model in-house.

3. Large model technology is still evolving rapidly, and Tencent will launch the more capable Hy4 model in the future. Factories can follow industry technical progress, map out intelligent transformation roadmaps aligned with their own production and design needs, and roll out relevant applications gradually.

The latest developments of Tencent's Hunyuan large model disclosed in this article provide the following industry insights for AI-related service providers:

1. The large model industry is seeing rapid technical iteration. Tencent's Hunyuan Hy3 validates the feasibility of the MoE architecture: it delivers performance matching flagship models with 2 to 5 times its parameter size using only 21 billion activated parameters, establishing a cost-effective high-performance technical path for the industry. The sector as a whole is moving toward lower costs, higher efficiency and faster commercialization.

2. Market demand for large models is now clearly defined, with enterprise clients showing strong demand for multimodal capabilities and large context windows. With Hy3 now commercially available, service providers can build industry-specific solutions on top of Tencent's mature Hunyuan capabilities to meet enterprise clients' AI empowerment needs and expand their business scope.

3. Tencent continues to scale up AI investment and will launch the more powerful Hy4 model in the future. Service providers can align with Tencent's technical iteration pace, prepare for integration with the new model in advance, and unlock more market opportunities in enterprise services.

The updates on Tencent's large model outlined in this article offer the following references for platforms looking to build out AI business:

1. The large model industry has now entered a phase of rapid commercial deployment. Weekly usage of Hy3 surged more than 68-fold compared to the previous generation immediately after launch, indicating very strong market demand. Platforms need to accelerate large model ecosystem building to keep up with industry development and meet the needs of users and developers.

2. Tencent's commercialization approach for large models provides a useful reference: it integrated the model internally into its existing office and coding products to boost the competitiveness of its own offerings, while externally acquiring new users and building an open ecosystem via API access, open-source licensing, and a free trial program for overseas users. This operational framework can serve as a blueprint for other platforms.

3. The large model industry is seeing extremely rapid technical iteration, with parameter scales and performance improving at a fast clip. Platforms building large model businesses need to reserve sufficient computing power and plan ecosystem development in advance to keep up with technical iteration and avoid the risk of falling behind technologically.

This article discloses the latest industry developments of Tencent's Hunyuan large model, providing the following key insights for large model industry research:

1. On industry investment: Leading Chinese tech companies are still drastically scaling up AI investments. Tencent's R&D spending increased 35% year-over-year in Q2 2026, and its capital expenditure rose 176% year-over-year, with investment focused on AI computing power procurement and large model R&D. This confirms that large models remain the core investment focus of the current AI industry, and the sector as a whole is in a phase of rapid upgrading.

2. On technical route research: Tencent's Hy3 adopts a MoE architecture, delivering performance matching large-parameter flagship models with a much smaller activated parameter count. This validates the efficiency of the MoE architecture for large model development and provides a new real-world test case for technical route research, with high research value.

3. On commercialization and industry trends: Tencent uses a hybrid commercial model of internal ecosystem deployment plus external API access and open-source commercial licensing, and acquires users via free trial programs. This commercialization model is worthy of in-depth research. Tencent's upcoming launch of the larger Hy4 model confirms that the large model industry is still in a phase of rapid iteration and upgrading, with continuous improvements to both technology and products.