Summary of Key Points
DeepSeek has released the new V4 Pro version and announced a significant increase in API prices (up to 12 times), while also implementing a peak-valley pricing system (higher prices during peak hours and lower prices during off-peak hours). Ordinary users can still experience the new model for free, but developers and enterprise customers who integrate the API into their products will face increased costs. The performance of the new model has reached the first tier of mainstream large models, although it is not yet completely leading; cost-effectiveness has become a major concern for developers.
1. How significant is the price increase?
The price increase far exceeds expectations:
- Specific increases: The cost for cached inputs has risen from 0.025 yuan per million tokens to 0.3 yuan during peak hours (a 12-fold increase) and 0.15 yuan during off-peak hours (a 6-fold increase); the cost for uncached inputs has risen from 3 yuan to 9 yuan during peak hours (a 3-fold increase) and 4.5 yuan during off-peak hours (a 1.5-fold increase); the cost for outputs has risen from 6 yuan to 27 yuan during peak hours (a 4.5-fold increase) and 13.5 yuan during off-peak hours (a 2.25-fold increase).
- Peak-valley timing: Peak hours are from 9:00 AM to 12:00 PM and 2:00 PM to 6:00 PM (these are the times when most people use the model for work), while the rest of the time is considered off-peak, with prices halved.
- Effective date: The new pricing will take effect on August 17, 2026 (developers have been given a nearly two-year buffer period).
In simple terms, it’s like taking a taxi during rush hours, which are more expensive, compared to off-peak times.
2. Who is affected by the price increase?
- Ordinary users: Using the V4 Pro in web or app interfaces in “expert mode” remains completely free and unaffected by the price increase.
- Developers/enterprise customers: Those who integrate DeepSeek models into their products (such as AI chatbots or intelligent customer service) will need to pay based on the number of tokens used. For example, a developer using 1 million tokens per day will spend an additional 0.275 yuan during peak hours, which amounts to over 100 yuan per year; for larger volumes, the cost can double.
3. Why is there a price increase?
The official explanation is “reasonable allocation of resources” – more people use the model during peak hours, leading to higher demand for computing power. However, possible underlying reasons include:
- The new V4 Pro model requires more advanced capabilities and thus higher development and computational costs (e.g., training more complex models with more GPUs).
- DeepSeek previously relied on low prices to gain market share; now, they may be raising prices to cover costs or generate profits.
- Industry trends: Many AI companies are increasing their prices (for example, OpenAI has also adjusted its pricing), so DeepSeek is following suit.
4. How does the new model perform?
The new V4 Pro offers several improvements:
- Enhanced Agent capabilities: It can better handle complex tasks, such as automatic planning and multi-step execution.
- Three levels of thinking intensity: Users can choose low for simple tasks, high for routine intelligent tasks, and max for more complex tasks (more intelligent but requiring more tokens).
- Evaluation rankings: It ranks 9th on Artificial Analysis and 8th on Arena.ai, second only to Kimi K3 Max among open-source models. However, these rankings may change as more users provide feedback.
In summary, the new model is not the most advanced, but it performs well among mainstream models, especially in the open-source community.
5. What do developers think?
The cost-effectiveness of the new model is a topic of debate, with many hoping for lower prices on Ascend chips (domestic AI chips):
- Criticism of the price increase: Some developers argue that the prices are now on par with similar models, making the Pro version less cost-effective, especially due to the high peak-hour rates.
- Hope for lower prices: There is a call for the widespread adoption of Ascend chips; if computing costs decrease, DeepSeek might lower its prices as well.
- Cost-effectiveness is key: Developers are concerned with using fewer tokens to complete more tasks. If the new model is more efficient, even if the price increases, the overall cost could be lower.
Conclusion
DeepSeek’s price increase marks a shift from a strategy of “low prices to gain market share” to one of “balancing costs and profitability.” The new model performs well, but its cost-effectiveness is questionable. While ordinary users are not affected, developers need to reassess their costs to decide whether to continue using it. Whether DeepSeek can retain customers in the long term depends on whether the new model can help them save on tokens – after all, an efficient and cost-effective model is what matters most.