Summary of Key Points
Recently, there’s been a counterintuitive development in the AI community: OpenAI announced that ChatGPT has exceeded 1 billion active users per week and made the moderately capable GPT-5.6 Luna available for free to everyone; meanwhile, DeepSeek, a Chinese large-scale model previously known for its aggressive pricing strategy, has decided to significantly increase its prices. This shift isn’t simply about who is generous or stingy, but rather about the strategic adjustments each company is making based on their own circumstances. OpenAI is using free access to attract users and build an ecosystem to generate revenue, while Chinese manufacturers are transitioning from a loss-making model to one that aims for profitability due to pressures related to computing power and costs.
1. OpenAI’s Free Offer: Not Charity, but a Strategy to Monopolize Access
Many people think OpenAI’s free offer is a charitable gesture, but there’s a clever strategy behind it:
- What’s included in the free version? GPT-5.6 Luna isn’t their most powerful model (the strongest ones are in the SOL series); it’s a middle-tier option that’s sufficient for everyday chatting, writing short essays, and simple translations. For more complex tasks like writing code, users still need to upgrade to the paid version.
- Why offer it for free? In the past 18 months, the cost of running large-scale models has dropped significantly (model optimizations and technological advancements have increased efficiency), meaning that serving one more user hardly incurs additional costs (marginal costs are very low). This is similar to Google Search: the search service is free, but it generates revenue through advertising and enterprise services.
- Where does the money come from? Although OpenAI is still in the red (losing $1.22 for every dollar earned in Q1 2026), its revenue sources have diversified. Advertising has generated $100 million in annual revenue, and corporate APIs contribute over 40% of its income, with Codex (a code development tool) attracting 5 million users per week. OpenAI’s goal is to become the “Google” of the AI era, making ChatGPT an essential part of people’s daily lives, and then charging for high-value services.
2. DeepSeek’s Price Hike: Not a Lack of Desire to Be Affordable, but Limited Computing Power
DeepSeek was previously known for offering its services at a fraction of OpenAI’s price, earning the nickname “the price killer.” Why is it raising prices now?
- Too many users, too little computing power: Its V4 Flash model is being used extensively, handling 7.22 trillion tokens per week, often leading to server lag and timeouts during peak times. The low cost attracted many casual users who didn’t need a stable service, taking up resources that could have been allocated to paying customers.
- Need to filter users: Raising prices is necessary to distinguish between those who use AI for fun and those willing to pay for high-quality services. This is similar to DouBao’s move to charge a monthly fee of 68 yuan; with 345 million active users, the daily cost of providing these services exceeds what they can earn from commissions.
- Financial and computational pressures: The domestic AI model financing environment has become more challenging, and computing power is subject to restrictions. The previous approach of spending money to acquire users no longer works; companies need to prove their profitability to survive.
3. Chinese Large-Scale Models: Moving from a “Price War” to a “Profitability War” – The End of the Cold Start Phase
The core advantage of Chinese models was their low cost. Why has that changed?
- The cold start phase is over: The initial goal was to make AI accessible to everyone, so they relied on low prices to gain market share. Now that users are accustomed to using AI, the focus is on making them willing to pay for better services (e.g., professional code generation and data analysis).
- Investors no longer support unprofitable models: They want to see returns on investment. If models don’t generate profits, who will invest? Therefore, the focus has shifted from the number of users to revenue.
- Not giving up on price competitiveness: Chinese companies are still offering competitive prices, but they’re more selective in their offerings, targeting paying customers instead of subsidizing everything.
4. Changing Competitive Logic: From “Who is Smarter” to “Who Cannot Do Without”
In the past, large-scale models were compared based on their performance (e.g., solving complex problems). Now, the gap between top models is narrowing, so focusing solely on performance is less meaningful.
- OpenAI’s strategy: OpenAI aims to make ChatGPT the default option for accessing AI, just as Google is for searching. By making ChatGPT a daily necessity, it can generate revenue through advertising and enterprise services even without direct user payments.
- Chinese companies’ strategy: Limited by computing power and funding, they cannot afford to offer free services on a wide scale like OpenAI. Instead, they focus on providing valuable, paid services in specialized areas to create a loyal user base.
- Mutual concerns: OpenAI is concerned about covering costs with free users; Chinese companies are worried about insufficient computing power affecting revenue generation. Both are seeking sustainable business models.
5. The Future of the AI Market: A Three-Tier Structure
This differentiation will lead to three distinct market segments:
- Lower tier: Free, basic services (chatting, information retrieval) that require minimal model capabilities and can be provided at no cost. Companies that charge for these will lose users (OpenAI’s free Luna aims to capture this segment).
- Middle tier: Paid professional services (code generation, legal assistance, medical consultations) that offer value and require a higher level of precision, for which users are willing to pay (DeepSeek and DouBao’s paid versions target this segment).
- Upper tier: Services where AI directly helps with tasks (business negotiations, code editing, auditing), where fees are based on the results achieved, not on the amount of data processed.
In summary, the era of free AI services is coming to an end. Only those services that can justify their costs will remain popular. OpenAI is focusing on gaining a dominant position in the market, while Chinese companies are aiming for higher-value, paid services. Whoever can establish a sustainable business model will succeed.