Summary of Key Points
2026 marks a watershed for the commercialization of large models: The past model-based revenue generation strategies, which relied on free C-side (consumer) chat services to attract users, are becoming increasingly ineffective due to high computational costs and difficulties in monetization. AI programming has become a core area where businesses and developers are willing to pay. Companies like ByteDance are shifting their focus from the C-side to B-side (business) AI programming and enterprise services. The AI programming market has divided into two main camps: high-end proprietary models (such as Anthropic and OpenAI) and low-cost, open-source platforms (like Zhipu and DeepSeek). ByteDance is positioned in the middle, attempting to enter this market with a "low-price + ecosystem" approach, but it faces challenges in both gaining B-side trust and demonstrating the capabilities of its models.
Detailed Analysis
1. Free C-side chat services are no longer profitable; the B-side is the real moneymaker
In the past, large model companies like ByteDance’s DouBao attracted users through free chat features, but each user interaction and generated content required significant computational resources (which were costly). For example, with 200 million daily users, DouBao’s daily revenue was less than one million, mainly from e-commerce commissions, while the computational costs amounted to tens of millions—meaning the more users there were, the greater the loss.
The B-side, however, is different: Businesses and developers are willing to pay for AI solutions that can actually solve their work problems. For instance, Anthropic’s Claude Code, an AI programming tool, was released in May last year and generated monthly revenue of 2.5 billion yuan by February this year, with half of that coming from corporate clients. The number of large customers spending over one million per year increased from 500 to 1,000 within just two months. As a result, companies are shifting their focus to the B-side in search of self-sustaining business models.
2. Why are people willing to pay for AI programming?
AI programming tools have evolved from being “toy items for programmers” to essential tools for enterprise development:
- For developers: 84% of developers are using or plan to use AI programming tools (according to a Stack Overflow survey in 2025), with nearly half using them daily. AI can automatically analyze requirements, read code, and run tests, significantly improving efficiency. Personal versions of these tools cost anywhere from a few dozen to over a hundred dollars per month, and more expensive professional versions are also popular (for example, Lin Ming personally subscribes to several such tools).
- For businesses: Fees are charged based on the number of users or usage tokens; as long as the tools are integrated into the development process, they generate stable cash flows. Corporate clients of Anthropic spend over one million per year, which constitutes a significant portion of the company’s revenue. In short, AI programming saves time and increases efficiency, making it a valuable investment.
3. ByteDance’s shift to the B-side: With confidence, but also with challenges
Where does ByteDance’s confidence come from?
- It has a solid foundation in the B-side: Its cloud services (Volcano Engine) and enterprise collaboration tools (Feishu) have been serving businesses for years. Its model service, Seedance, generates annual revenue of 14.3 billion yuan with a gross margin of 70%, nearly covering DouBao’s computational costs alone.
- Low-price + compatibility strategy: ByteDace’s Coding Plan is affordable and compatible with mainstream development tools like Claude Code and Cursor, allowing developers to switch from using other platforms without changing their work processes, thus reducing migration costs.
What are the challenges?
The B-side requires stability and trust; while free or subsidized C-side strategies worked well in the past, the B-side demands reliability and consistency. ByteDance’s culture of rapid experimentation may make businesses wary of product instability (e.g., AI-generated code that needs significant modifications before deployment). Additionally, developers are still heavily inclined towards leading platforms like Claude and ChatGPT, making it difficult for ByteDance to attract them.
4. The players in the AI programming market: Two camps competing for market share
The market is divided into two main camps:
- High-end models (Anthropic, OpenAI): These companies sell “AI employees” that can handle complex tasks, earning revenue from enterprise subscriptions. For example, Claude’s corporate clients are willing to pay millions per year for its advanced capabilities. OpenAI’s Codex saw a sevenfold increase in weekly active users within half a year, reaching over 5 million.
- Low-cost platforms (Zhipu, DeepSeek): These companies offer model-based solutions, competing through API calls and open-source models at lower prices. Zhipu’s Coding Plan costs one-seventh of Claude’s price, while DeepSeek appeals to users with its affordability and suitability for simpler tasks.
ByteDance is trying to position itself in the middle, aiming to develop high-end Agent systems that can deliver stable results, but it faces challenges in both competing with established players in terms of user adoption and cost-effectiveness.
5. What will determine future competition?
The gap in model capabilities is narrowing (for example, Zhipu’s GLM5.2 is comparable to Claude’s). The key factors for success will be:
- Cost control: Computational resources are expensive, so it’s crucial to manage costs effectively. Subscription models with usage limits and API fees based on tokens can help prevent excessive use and losses (Zhipu had to limit sales due to high demand).
- Product stability: The ability to handle complex tasks and provide timely support in case of issues is essential for B-side businesses.
- Developer trust: Users will choose the tools they find most useful, with no loyalty to a particular platform. For example, Lin Ming spends most of his budget on Claude/ChatGPT for complex tasks and uses DeepSeek for simpler ones. To retain users, it’s necessary to continuously provide a good experience.
This competition is intense, and the winner will be the one that can sustain itself longer and effectively solve real business problems.
Conclusion
The commercialization of large models has shifted from focusing on traffic to generating revenue through practical applications. AI programming is the most promising area for monetization. ByteDance’s strategy reflects this trend, indicating that future winners will need not only powerful models but also the ability to convert computational costs into actual productivity and gain the trust of businesses and developers. This battle has just begun, and it’s still uncertain who will emerge as the winner, but the direction is clear: The B-side represents the true future of large models.