Summary of Key Points
DeepSeek’s launch of an ultra-low-cost large model has sparked intense discussion about the “cutting-off line” it represents. However, this model will not directly eliminate top-tier closed-source models like OpenAI and Anthropic. Instead, it will first target the mid-range models, which are mediocre in capability but excessively expensive. Additionally, low-cost models will enable more standardized white-collar workers to overcome the cost barriers of automation, ultimately changing the pricing logic of the model industry and the nature of the workplace.
Analysis in Detail
1. Who Will Be the First Target of the Cutting-Off Line? Mid-Range Models
Mid-range models are those that are neither as powerful as top-tier models (such as OpenAI GPT-4) nor as affordable as low-cost models like DeepSeek V4 Flash. Companies currently use a “model routing” strategy, utilizing expensive models for complex tasks (e.g., strategic planning, critical decision-making) and cheaper models for simpler tasks (e.g., batch summarization, format conversion, data entry). Mid-range models lack the irreplaceability of top-tier models and the cost advantages of low-cost models. As a result, customers naturally question: “If both can perform the same task, why pay several times more?” Therefore, mid-range models will be the first to be phased out.
For example, data from OpenRouter shows that one month after the release of DeepSeek V4 Flash, it accounted for 70% of the traffic in its agentic (intelligent agent) services—indicating that companies are beginning to assign more repetitive tasks to lower-cost models, squeezing the market space for mid-range models.
2. OpenAI and Anthropic: Short-Term Challenges, but Not Certain Long-Term Losses
In the short term, DeepSeek’s low-price strategy does put pressure on these models. For instance, OpenAI’s Luna model saw its price drop by 80% three weeks after its launch, and Anthropic’s revenue growth has slowed down (although it is still growing overall). However, in the long run, they have two key advantages:
- Cutting-edge Capabilities: They possess advanced capabilities for complex reasoning, multi-step data analysis, and enterprise-level reliability (e.g., stability and data security), which low-cost models cannot yet match.
- Ecosystem Advantages: OpenAI has a wide range of products and tools (e.g., code interpreters), while Anthropic has a strong presence in enterprise services. Customers are willing to pay for these additional benefits and convenience.
As long as they continue to maintain their leading edge in capabilities, they will be able to retain control over the most critical and error-prone tasks, making them less susceptible to replacement.
3. DeepSeek’s Business Model: Open Source Isn’t Free; Profit Comes from Efficiency
Many wonder how DeepSeek makes money since its models are open source. The logic is simple:
- Open source is a strategy to spread adoption: By making the models available to more people, DeepSeek aims to build user loyalty.
- Profit comes from deployment efficiency: While open-source models can be downloaded, large-scale, stable operations require additional costs (GPUs, data centers, scheduling systems, etc.). DeepSeek optimizes its model architecture and tools (e.g., TileLang) and adapts them to Ascend hardware to minimize these expenses. Its pricing strategy is designed to ensure a reasonable return on investment within 10 months of deployment, focusing on high volume and managed services rather than high margins.
In other words, if you need to run the model yourself, it might cost $10; DeepSeek can do it for only $3—this is where its profit lies.
4. The Ultimate Impact of the Cutting-Off Line: Redrawing the Price Threshold for White-Collar Work
The real impact of this trend is not the elimination of model companies but a reshaping of the workplace. Many white-collar tasks (e.g., editing documents, researching information, writing templates, providing standardized customer service) were previously too costly to automate. Now that models can perform these tasks for just a few cents per million characters, companies realize that AI is more cost-effective. Repetitive and standardized tasks will be the first to be automated.
Not all white-collar workers will lose their jobs, but the value of such roles may decrease, potentially leading to lower salaries or replacement by AI.
5. The Future of the Industry: A Coexistence of Closed-Source and Open-Source Models
In the short to medium term, the industry will not see a clear winner between closed-source and open-source models:
- Closed-Source Models in the U.S.: They still lead in cutting-edge capabilities (e.g., complex reasoning with GPT-4).
- Open-Source Models in China: They are rapidly catching up in terms of cost-effectiveness and specific tasks (e.g., Kimi K3 in long-text processing).
The two types of models will coexist, with closed-source models handling high-end tasks and open-source models covering more general use cases. The value will shift from selling model licenses to providing services such as deployment and customized solutions for enterprises.
Final Conclusion
The so-called “cutting-off line” is not a death knell for any company but rather an accelerator for the widespread adoption of AI. It makes models more affordable and automation more feasible, forcing model companies to improve their capabilities and workers to develop indispensable skills (e.g., complex decision-making, interpersonal coordination). For businesses, this means cost reduction and efficiency improvement; for workers, it represents a need to adapt or risk being replaced by AI.