字节跳动重组Seed大模型团队 筹划5万亿参数模型

This article reveals the latest internal restructuring and R&D plans of ByteDance's Seed large language model (LLM) division. Key takeaways are as follows:

1. Restructuring details: Seed completed internal restructuring in August 2026. The original team was split into four first-level departments: pre-training data, Horizon RL, product post-training for workplace, and product post-training for chat. The two post-training departments have clear division of labor, focusing on B-end business agent R&D and C-end consumer dialogue model development respectively.

2. Current achievements: Founded in 2023, Seed is ByteDance's core LLM R&D team. It has launched multiple open-source models and technical achievements that power over 50 application scenarios. Its domestic AI assistant Doubao reached 172 million monthly active users (MAUs) in October 2025, covering multi-terminal entry points. ByteDance's overseas AI assistant Dola, after rebranding, has surpassed 10 million daily active users (DAUs) and once topped the app store download charts in multiple countries.

3. Future plans: ByteDance is discussing the development of an LLM with over 5 trillion parameters. The plan is still in its early stage and has not been officially announced as a product.

This article discloses ByteDance's latest layout in the AI large model space and maps out current trends in the consumer AI market, offering insights for brands looking to integrate AI into their strategies. Key takeaways are as follows:

1. Consumer trends: C-end AI assistants have already built massive user bases—Doubao boasts over 170 million MAUs, while overseas counterpart Dola has more than 10 million DAUs. This demonstrates that consumer demand for AI assistants is fully activated, and AI assistants have become important new traffic entries. Brands can explore new marketing channels by integrating with these AI assistants.

2. Commercialization momentum: By splitting its team to refine division of labor and advancing R&D for both B-end business applications and C-end consumer products simultaneously, ByteDance is accelerating LLM commercialization. As LLM capabilities mature, they will increasingly empower brands across multiple functions including product development and user operations.

3. Future outlook: ByteDance has started early-stage planning for a 5 trillion-parameter large model, which will further upgrade AI capabilities down the line. Brands that lay out AI applications early can capture user mindshare and gain a competitive edge.

This article discloses the latest industry trends and opportunities for sellers operating within ByteDance's ecosystem and pursuing AI-related business. Key takeaways are as follows:

1. Ecosystem changes: ByteDance's restructuring of its LLM team—splitting into function- and scenario-specific departments to clarify division of labor—signals the company is accelerating LLM commercialization. ByteDance's AI ecosystem will mature rapidly, providing sellers with more sophisticated AI tools and stronger traffic support.

2. Market opportunities: AI assistants have achieved massive user growth in both domestic Chinese and overseas markets, with Doubao hitting 172 million MAUs in China and Dola surpassing 10 million DAUs abroad. AI assistants have emerged as a new high-growth market. Sellers can explore traffic opportunities by integrating with AI assistants and tap into new demand across AI-related scenarios.

3. Guidance for future布局: ByteDance is planning an ultra-large-scale parameter LLM, and future AI capability upgrades will unlock new business model opportunities. Sellers can track relevant developments early, adjust business strategies ahead of the curve, and capture the next wave of growth.

ByteDance's latest large model layout offers insights for factories pursuing digital transformation and exploring new business opportunities. Key takeaways are as follows:

1. Implications for product development and design: Large model technology is now mature and supports more than 50 application scenarios. Factories can leverage mature LLM capabilities to optimize product design and R&D workflows, cut R&D costs, improve efficiency, and launch new user-aligned products faster.

2. Business opportunities: ByteDance has established a dedicated team focused on workplace business applications of large models, which signals more AI tools tailored for B-end production will be launched in the future. Factories can watch for the release of these tools and use AI to drive digital upgrades in production and management processes.

3. Transformation direction: With ByteDance planning a 5 trillion-parameter ultra-large model, AI's comprehension and generation capabilities will improve further, and will empower more links including factory production and supply chain management. Factories that advance digital transformation and integrate AI capabilities early can build a competitive advantage ahead of peers.

This article discloses the latest industry developments in the large model space, providing guidance on business direction for AI-related service providers. Key takeaways are as follows:

1. Industry trends: The large model industry has moved beyond early-stage R&D and entered a phase of specialized, scenario-specific deployment. Industry leader ByteDance has already split its R&D team by function and scenario, indicating that industry division of labor is becoming increasingly granular, and demand for professional solutions tailored to specific scenarios is growing rapidly.

2. Customer pain points and demand: The industry continues to pursue performance upgrades, and leading players have already begun planning ultra-large parameter models with over one trillion parameters. This means market demand for higher LLM performance is continuously rising, creating substantial market opportunities for supporting services around ultra-large model R&D, including pre-training data processing, model optimization, and scenario adaptation.

3. Market opportunities: ByteDance's AI assistants have already acquired massive user bases in both Chinese and overseas markets, and a large number of enterprises have demand for accessing ByteDance's LLM capabilities and deploying AI applications. Service providers can build supporting services around ByteDance's LLM ecosystem to meet the deployment needs of diverse clients.

ByteDance's approach to large model development offers valuable reference for platform companies building their own LLM strategies. Key takeaways are as follows:

1. Organizational restructuring reference: ByteDance split its original LLM team into four first-level departments segmented by function and scenario, with clear division of labor that improves R&D and deployment efficiency. This restructuring model is well suited for platform companies building LLM businesses, as it adapts to the needs of moving from R&D to commercialization.

2. Business layout direction: Consumer demand for C-end AI assistants has been validated, with significant room for growth in both domestic and overseas markets. Platform companies entering the AI assistant space can adopt a dual domestic-overseas market strategy to capture user traffic.

3. R&D guidance: Ultra-large parameter large models are a key R&D direction for the future. Platform companies can build relevant technical reserves early, but should note that ultra-large models are still in the early stage. It is important to control R&D investment reasonably to avoid risks from going "all in" too early.

This article discloses the latest developments in large model business at a leading industry player, providing new first-hand source material for AI industry research. Key takeaways are as follows:

1. New industry organizational trends: A leading internet company has restructured its organizational architecture for its LLM business, splitting R&D departments by function and application scenario. This indicates that the large model industry has transitioned from technology exploration to commercial deployment, and organizational adjustments are being made to adapt to the needs of this new development stage—this is an important new feature of industry growth.

2. New technology R&D trends: The leading company has started early-stage planning for a 5 trillion-parameter ultra-large model, confirming that the parameter race in the LLM industry is ongoing, technological upgrading is accelerating, and ultra-large models are currently a key R&D direction for the industry.

3. New commercialization characteristics: ByteDance has built a dual AI assistant product matrix with Doubao for the Chinese market and Dola for overseas, both of which have achieved massive user scale. This validates the commercial viability of C-end LLM assistants. At the same time, the company is advancing B-end commercial application deployment in parallel. This dual C-end and B-end commercialization path provides a new case for research on LLM commercialization.