Summary of Key Points
DeepSeek recently announced a significant increase in its API service pricing, citing the previously low prices as being commercially unsustainable. The company also clarified rumors about “Fantang subsidizing DeepSeek through stock trading,” emphasizing that it has independently raised funds and plans to go public. It acknowledged several shortcomings, including the lack of a subscription model, limited multimodal capabilities, and slow update speeds. Finally, it noted that the competition in the domestic large-model industry remains intense, with no clear winner yet.
1. Price Hikes Are Not a Form of Exploitation, but a Business Necessity
The price increase is substantial, not just a minor 10%-20% rise. Why? Simply put, the prices were too low before! Compared to other domestic large models (such as GLM), DeepSeek’s API fees were already competitive, even cheaper than those of overseas models like GPT and Claude. Developers felt that even doubling the prices would still make it competitive. From a business perspective, the cost of developing large models (both in terms of computing power and manpower) is high, and maintaining low prices was unsustainable. Moreover, DeepSeek’s V4 Flash version has received positive feedback from users, making the price hike more acceptable.
2. “Fantang Subsidizing DeepSeek Through Stock Trading”? Don’t Believe That
Some online claims suggest that Fantang uses profits from stock trading to subsidize DeepSeek in an attempt to undercut overseas models. This is nonsense. DeepSeek became independent when its V3 version was released, and it now has its own funding sources and plans for listing, aiming to be self-sufficient rather than relying on others. Additionally, the amount of money Fantang earns from stock trading is not enough to challenge giants like OpenAI or Anthropic, which have substantial annual recurring revenues.
3. Three Major Weaknesses That Users Care About
Despite its competitive pricing, DeepSeek has several notable drawbacks:
- No Subscription Model: Currently, it only charges based on the amount of usage (tokens), without monthly subscription options. This can be inconvenient for individual developers or small teams that need a fixed budget (other models like GLM and Kimi offer monthly subscriptions).
- Limited Multimodal Capabilities: It can only process text and cannot recognize images or videos, which is a major limitation for applications requiring image analysis or video content generation.
- Slow Update Speed: Its updates are slower than some other domestic models, which may disappoint users who prefer the latest features.
4. Price Hikes Make the Industry More Balanced
The previous low prices put significant pressure on other domestic large models, as users tended to choose the cheaper options. With the price hike, DeepSeek’s competitiveness is more balanced with that of its peers, creating a fairer market environment where all companies can thrive. The demand for AI tools continues to grow, and as long as the products are useful, there will be enough customers.
5. Domestic Large Models Are Still Competing Intensively; No Clear Winner Yet
This year, domestic large models have been competing fiercely: GLM-5.2 dominated in June due to its coding capabilities, followed by Kimi K3 in July, and then DeepSeek V4 Flash gained attention. Recently, Qwen 3.8 was praised for being comparable to Claude. Overseas models like GPT and Grok are also making progress. This indicates that the competition is far from over, and it will take at least 2-3 years to see a stable landscape. It’s too early to determine a clear winner.
In summary, DeepSeek’s price hike is a normal business decision that addresses its financial needs and promotes a more balanced industry. For users, the increased cost may be a downside, but as long as the product quality remains high, it will still be valuable. For the industry as a whole, it marks a transition from a phase of reckless spending to one of sustainable profitability.