Pricing logic in the domestic large model industry is reaching a pivotal turning point.

DeepSeek announced on August 6th that it plans to raise prices for its API services across the board, noting the expected increase would be substantial, with the final details to be confirmed later. While the company has not yet disclosed the specific rate or timeline for the adjustment, the market widely believes this signals a departure from the industry's previous strategy of using low prices to capture market share. The sector is now entering a new cycle centered on cost, efficiency, and commercialization.

This price hike is driven by the dual pressures of surging demand for AI applications and rising computing costs. The emergence of advanced applications, such as AI Agents, has led to a massive increase in model call volumes, challenging the previous model of stimulating demand through low prices.

Howaring Demand Forces DeepSeek to Re-evaluate Pricing

The core reason for this price increase is that the demand for AI applications has grown faster than anticipated.

Data from the open-source AI Agent tool OpenCode shows that the daily token call volume for DeepSeek V4 Flash on its platform reached 8 trillion, with 5 trillion from free credits and 3 trillion from paid plans. For comparison, the widely-used model routing platform OpenRouter, which hosts over 400 models, has a total daily token call volume of about 6.6 trillion.

This means that DeepSeek's single model, on just one platform, processes more calls than the entire daily average of a comprehensive model aggregation platform.

Data from the Vercel platform also indicates that DeepSeek's token processing volume continues to climb. The V4 Flash model handles approximately 5.3 trillion tokens per week, making it one of the most-used models on the platform. As AI Agents become more widespread, the pattern of model calls is changing. Where a chatbot interaction might have consumed just hundreds or thousands of tokens, an Agent performing a complex task requires continuous model calls and tool accesses, potentially consuming tens or even hundreds of times more tokens than a traditional conversation.

With this explosion in demand, extremely low prices are becoming a significant new cost pressure.

DeepSeek has stated that the price increase is mainly due to three factors: the rapid growth of agent applications leading to a surge in token consumption; rising computing costs due to restricted supply of high-end AI chips; and the industry's shift from a model of using subsidies to gain scale towards a sustainable business model.

From "Price Killer" to Commercial Pricing

DeepSeek's shift in pricing strategy is not sudden.

At the end of April, DeepSeek launched a limited-time discount for its V4-Pro API, which was then made permanent in May, reducing the output token price by 75% from the original rate. At that time, DeepSeek V4-Pro's output price was about $0.87 per million tokens, far lower than leading overseas models, earning it the market nickname "Price Killer."

This strategy triggered a wave of collective price cuts among domestic large model companies. Companies like Xiaomi, ByteDance, and Tencent Cloud successively adjusted their pricing systems, with some model prices dropping by over 90%, plunging the domestic market into intense price competition.

However, at the end of June, DeepSeek began to alter its approach. The company introduced a peak-valley pricing mechanism for the official V4 version, doubling API prices during peak workday hours. At the time, the market saw this as a way for DeepSeek to manage demand and optimize computing resource utilization.

This current comprehensive price increase, however, signifies that DeepSeek is moving from "managing traffic" to "redefining its price."

At a previous investor meeting, DeepSeek founder Liang Wenfeng stated that API pricing should be based on reasonably recovering equipment costs, with a goal of "buying a batch of equipment and recovering the cost in ten months." He also pointed out that user demand is "almost inelastic" within the current price range, meaning that even if prices rise, the change in token consumption would be limited.

Computing Power Shortage Makes Low-Price Model Unsustainable

Behind DeepSeek's price adjustment lies the intensifying supply-demand imbalance across the entire AI computing power industry chain.

With the rapid growth in demand for large model training and inference, resources like GPUs, servers, and storage remain under constant pressure. In the AI Agent era, inference demands have shifted from simple Q&A to long-chain task execution, vastly increasing computing power consumption.

Goldman Sachs previously noted that DeepSeek's implementation of peak-valley pricing does not indicate weak demand, but rather reflects strong domestic demand for AI models and tightening computing resources. As AI applications enter a stage of scaling, maintaining extremely low API prices over the long term will be difficult to sustain against ever-increasing computing investments.

In fact, over the past six months, the domestic large model industry has shown collective signs of adjustment.

Companies including Alibaba, ByteDance, Xiaomi, Zhipu AI, and Tencent have all adjusted their product pricing or plan structures, with some low-price offerings beginning to shrink. Zhipu AI raised its API call prices in the first quarter of this year, yet call volumes continued to grow rapidly. Kimi also had to pause new user subscriptions due to request volumes significantly exceeding expectations.

The industry is shifting from "burning money to acquire users" to "scaling commercially to generate revenue."

Domestic Large Models Enter the Phase of Profitability Validation

DeepSeek's price increase also sends a more important signal: domestic large models are moving from technological competition to commercial competition.

Over the past year, low-cost or even free model services helped domestic large models rapidly expand their user base. However, as call volumes have exploded, the costs of computing power, server investment, and operational pressure have become apparent.

Previously, the market's focus was on who could offer the cheapest model. The future competitive battleground will shift to who can deliver higher performance and more stable services at a lower cost.

For developers, DeepSeek's price increase means the cost advantage for AI applications is diminishing. However, it also signals that the industry is beginning to move towards a healthier commercial cycle.